Faculty of

Information Technology
6HUYHU7HFKQRORJ\
Year  Semester 
Registered with the Department of Higher Education as a Private Higher Education Institution under the Higher Education Act, 1997.
Registration Certificate No. 2000/HE07/008

FACULTY OF INFORMATION TECHNOLOGY

LEARNER GUIDE
MODULE: 6(59(57(&+12/2*< 6 (2ND SEMESTER)

PREPARED ON BEHALF OF
RICHFIELD GRADUATE INSTITUTE OF TECHNOLOGY (PTY) LTD

GRADUATE INDTITUTE OF TECHNOLOGY (PTY) LTD

Registration Number: 2000/000757/07

All rights reserved; no part of this publication may be reproduced in any form or by any
means, including photocopying machines, without the written permission of the
Institution.

2
 SECTION B

Registered with the Department of Higher Education as a Private Higher Education Institution under the Higher Education Act, 1997.
Registration Certificate No. 2000/HE07/008

FCAULTY OF MEDIA INFORMATION AND COMMUNICATION TECHNOLOGY

HIGHER CERTIFICATE IN SYSTEMS ENGINEERING

LEARNER GUIDE

MODULE: SERVER TECHNOLOGY 600(2ND SEMESTER)

TOPIC 1 : DEFINING A SERVER

TOPIC 2 : MOTHERBOERD ARCHITECTURE, PROCESSORS,
MEMORY, AND BIOS
TOPIC 3 : SERVER POWER AND RACK INSTALLATION
TOPIC 4 : HARD DISK INTERFACES AND RAID
TOPIC 5 : SERVER UPGRADES
TOPIC 6 : PLAN AND IMPLEMENT SERVER AVAILABILITY
AND SCALABILITY
TOPIC 7 : HARDWARE AND SOFTWARE MANAGEMENT

PRESCRIBED & RECOMMENDED MATERIAL

Prescribed Material:

Roberta Bragg: Designing Security for a Microsoft Windows Server 2003 Network, 1st Edition.

The purchasing of prescribed books is for the learners own account and is compulsory for all learners. This
guide will have limited value if not accompanied by the prescribed text books.

12
 HIGHER CERTIFICATE IN
TOPIC 1 : DEFINING A SERVER
SYSTEMS ENGINEERING
1.1 Does your network need a server
Lecture 6
1.2 Client Server
1.3 Server functions and Benefits Lecture 7
1.4 Processors & memory
Lecture 8
1.5 Mainframe Computers
1.6 Server Planning Lecture 9
1.7 Server placement
1.8 Intra Site Server placement Lecture 10
1.9 Creating the network Diagram
1.10 Planning Physical site readiness Lecture 11
TOPIC 2 : MOTHERBOARD ARCHITECTURE, PROCESSORS,
MEMORY, AND BIOS
2.1 Mother Board Architecture Lecture 12
2.2 Processors Lecture 13
2.3 Memory Lecture 14
2.4 CMOS, BIOS and POST Lecture 15
TOPIC 3 : SERVER POWER AND RACK INSTALLATION
3.1 Server Power Lecture 16
3.2 The RACK
Lecture 17 – 18
3.3 RACK installation Tips
TOPIC 4 : HARD DISK INTERFACES AND RAID
4.1 Hard Disks Lecture 19
4.2 File Systems Lecture 20 – 21
4.3 The SCSI Interface Lecture 22 – 23
4.4 Fiber Channel & RAID Lecture 24
TOPIC 5 : SERVER UPGRADES
5.1 Preparing for a server upgrade Lecture 25
5.2 Upgrading the processor and memory Lecture 26
5.3 Upgrading the BIOS, upgrading power Lecture 27
TOPIC 6 : PLAN AND IMPLEMENT SERVER AVAILABILTY AND
SCALABILITY
6.1 Availability & scalability in WINDOWS Lecture 28
6.2 Install & configure server clusters Lecture 29
6.3 Server Cluster Concepts Lecture 30
6.4 Network load Balancing Lecture 31
TOPIC 7 : HARDWARE AND SOFTWARE MANAGEMENT
7.1 Hardware Management Lecture 32 – 34
7.2 Software Management Lecture 35
7.3 Application Management Lecture 36
13
 TOPIC1.
DEFINING A SERVER
_____________________________________________________________

LEARNING OUTCOMES
After studying this topic you should be able to:
• Discuss the difference between peer-to-peer and client-server networking models.
• Identify server functions and benefits.
• Identify three main types of servers.
• Determine the most appropriate server solution for a given business purpose, plan for user
demands on the server, and consider interoperability among operating systems.
• Optimize server placement and diagram server plans.
• Plan the server environment and physical site readiness. Implement sound physical server
security practices.

1.1 DOES YOUR NETWORK NEED A SERVER?

Before you look into planning and installing a server, consider if you really need a Server. This might seem
like an odd statement to place in the first chapter of a book on servers, but it is a legitimate one. If your
organization is small (perhaps 12 or fewer workgroup computers), users only occasionally share printers,
files, or applications, and security is not a major concern, then it is possible to network all the computers
together without a server. For example, users in a small, family-run construction company probably
already trust one another and only need to print invoices or checks to a shared printer from time to time.
The company does not require a database, and the construction managers create job estimates on their
own laptop PCs, which only occasionally connect to the network. In this case, it is of little use to implement
a server.

Conversely, most organizations (even small ones) can benefit from adding one or more servers to the
network. Some of the benefits include enhanced security, improved performance, centralized file storage,
centralized administration, and a central location from which to run applications. For example, the owner
of a small real-estate office installs a high speed Internet connection so that real-estate agents can
research land values and receive email from clients. Also, the owner wants each agent to enter sales
records into a company database. A server could include special software to protect office users and the
company database from malicious Internet users and provide a central location to store the company
database.

After determining the need for a server, you must also determine “how much” server you need. Will the
server support dozens, hundreds, or thousands of users? Do you require extremely high performance?
Some servers offer features and performance comparable to common desktop personal computers, some
are extremely powerful and cost hundreds of thousands of dollars, and others fall somewhere in the
middle.

14
Comparing peer-to-peer and client-server models
A network is a collection of two or more computers connected with wired transmission media such as
network cable or with wireless radio or infrared signals, and it usually includes other devices such as
printers. A network device is any device connected to the network for purposes of communicating with
other network devices. (A network device is also known as a host in most networks.) Cumulatively, the
network devices form what is known as a local area network (LAN)—a collection of computers in close
proximity to one another on a single network. A LAN includes a single cable to which all computers connect
or, more commonly, each device has its own network cable, each of which converges to a central hub or
switch.

You can use a network cable to join multiple hubs together, but the entire collection of hubs still
constitutes a single network. A LAN can be one of two basic networking models: peer-to- peer or
clientserver. Expanding upon a LAN, a wide area network (WAN) involves multiple, geographically distant
LANs connected to one another across relatively great distances. For example, an organization’s LAN in
Seattle connected to the same organization’s LAN in Phoenix constitutes a WAN.

Peer-to-peer
A network does not necessarily require a server. A peer-to-peer network is a network of computers with
no logon server to verify the identity of users. (Although a peer-to-peer network might include a file server
that stores files for the users, this is seldom the case.) This model is called a peer-to-peer network because
each network device has an equal (peer) level of authority.

Computers in a peer-to-peer network are usually common desktop computers, otherwise known as
workstations, and are generally equipped with only enough hardware to service the needs of a single user.
Most peer-to-peer networks use the file and printer sharing capability of Microsoft Windows
95/98/ME/NT/2000 to share network resources. A network resource (sometimes called only a “resource”)
is an object that users can access across the network. Common examples of network resources include
printers, files, and folders.

Although the file and printer sharing capability of a peer-to-peer network can service a small network, it
can also limit network growth. For example, Microsoft Windows 95, 98, NT Workstation, and 2000
Professional limits inbound concurrent network connections to 10. This means that if 10 users are already
accessing a folder, an additional attempt to access the folder will fail because the limit of connections has
already been reached. This limitation restricts the size and growth capacity of peer-topeer networks. Peer-
to-peer networks function best when individual users seldom access one another’s files. If multiple users
share a workstation, confusion might result.

Peer-to-peer networks also lack file security. Windows 95/98/ME share folders (and hence the files
contained in the folders) with password protection. A password can be easily compromised and, at best,
is only as secure as the operating system. Most desktop client operating systems are designed with
minimal security. For example, any passerby can boot up a Windows desktop client and, whether logging
on to the network or not, access local resources on the entire local hard disk.

Windows NT or 2000 workstations in a peer-to-peer configuration, on the other hand, require you to
create a user account that has a username and password in order to log on locally. Access to folders or
individual files can be specified for designated users. If a user attempting to access your shared files across
the network does not have valid credentials for the shared files, he or she is denied access. Although

15
Windows NT or 2000 is inherently more secure than Windows 95/98/ME password security, a peer-to-
peer network requires you to store user accounts on each workstation.

For example, suppose you, Shelly, and Karl all have Windows NT 4.0 workstations in a Peer-to-peer
network. You want to share a folder with Shelly, and Karl also wants to share a folder with Shelly. Both
you and Karl must create a user account for Shelly on your respective workstations, and Shelly’s username
and password must be exactly the same. If Karl changes the password for Shelly’s user account on his
machine and you don’t, then Shelly’s credentials will be inconsistent and she will not be able to access
resources where her password is different from the one she used to log on (see Exhibit 11). The root of
this problem is that there is no centralized location in which to store user accounts, a significant weakness
of the peer-to-peer network. Managing users and resources without centralized user account
management can quickly become a time-consuming nightmare.

Exhibit 1-1: Inconsistent user accounts can cause access problems

A peer-to-peer network is simple to configure, but it is limited in terms of expandability, features, services,
and security. Many small networks are peer-to-peer networks because the organizations they serve do
not require (or cannot afford) the benefits of a server, or there is insufficient technical expertise to
administer a server. A peer-to-peer network:
• Usually involves a small number of computers (around one dozen or less)
• Has limited growth potential
• Has decentralized file management, user accounts, and overall management
• Offers minimal security
• Is simple to configure
• Is typically the least expensive option

1.2 CLIENT-SERVER

The client-server networking model has most of the benefits of the peer-to-peer network model and
potentially none of its weaknesses. The client-server network begins with a LAN and one or more servers,
but it can also encompass a more complicated network configuration such as a WAN.

16
A server usually possesses more processing power, RAM, and hard disk capacity than workstation
computers on the LAN. The server also has a server network operating system (NOS) such as Microsoft
Windows NT or 2000, Linux, IBM OS/2, or Novell NetWare. A server running a NOS provides file and printer
sharing, centralized file storage, administration, security, services, and significantly more stability than
desktop operating systems.
At a basic level, simply installing a server into an existing peer-to-peer network changes the network model
to a client-server network. Workstations in the client-server network request and receive services (access
to files, printers, or applications, for example) from the server—hence the term client. The server serves
the client—hence the term server.

Although users at the workstations can save files on their local hard disks, it may be more practical to save
files on the server so that all users have a central location to access files. A server in the role of storing
files on behalf of network clients is commonly referred to as a file server. A file server also offers other
advantages, including:
• Version control: Using a file server helps to avoid the situation in which differing versions of a file
are saved on various workstations.
• Backup: Network administrators can back up data files centrally on the file server as opposed to
hunting for and individually backing up the data on each user’s workstation.
• Security: Network administrators can centrally audit file resources to see if unauthorized persons
have tried to gain access to them. Also, administrators can configure security settings on the files
instead of users, improving the chances that sensitive files will be well protected.
• Availability: The file server is usually placed in a well-connected location to increase the availability
of file resources. Because of the stability of the NOS and the use of redundancy, the server is less
likely to be subject to outages or downtime. (Redundancy is the ability to continue providing
service when something fails. For example, if a hard disk fails, a redundant hard disk can continue
to store and serve files.) In addition, the server remains on 24/7, so it is always available to clients,
as opposed to a computer in a peer-to-peer network in which resources are only available when
the user turns on the computer.
• Integrity: NOS file systems protect the integrity of their files by using special logs and error
correction that can repair a damaged file on the fly or issue an alert to an administrator.

Because a NOS is capable of hundreds or thousands of simultaneous connections, servers can significantly
increase the size and growth potential of a network. Some servers can even handle millions of
simultaneous connections provided the server hardware and network bandwidth can match the demand.
A server in the client-server network model:
• Possesses more processing power, RAM, and hard disk capacity than typical workstations
• Uses a NOS such as Microsoft Windows NT, Windows 2000, Linux, IBM OS/2, or Novell NetWare
• Provides a central file storage location
• Is capable of many more simultaneous connections than a workstation
• Offers security features such as logon authentication
• Provides centralized administration

All these benefits are the result of server implementation in the network. In addition to these features,
servers offer several other functions and benefits.

17
1.3 SERVER FUNCTIONS AND BENEFITS

The server’s impact on the network varies depending upon which of its many features you choose to
implement. Even adding a single server to a peer-to-peer network adds benefits to the network in terms
of services, security, performance, storage capabilities, access to applications, and centralized
management. These features add up to more cost effectiveness, efficiency, and productivity.

Services
A service is a function of the NOS that provides various server functions and benefits to the network. For
example, a company Web server could host many services. When a user accesses the Web site by typing
a Uniform Resource Locator (URL) into a Web browser, the Domain Name System (DNS) service translates
the URL into a unique number that identifies the server that contains the Web pages. When the user views
the Web page, he or she might want to leave a message for someone at the site by clicking a link that
sends an e-mail message. The e-mail functionality requires an e-mail service. If the user purchases
products from the Web site and wants to view his or her account information, a service checks the user’s
username and password so that only that user can view the account. Yet another service could filter and
monitor all traffic from the user’s network to and from the Internet for security purposes. In most
networks, these services are probably distributed among several servers for purposes of redundancy and
performance.

Security
Protecting the network against unauthorized access is a significant challenge. Though you want to grant
network access to users, you do not want to allow them free reign on the network and its resources.
Conversely, too many restrictions will be counterproductive. The administrator must permit a level of
access that is only as much as the user needs.

Authentication
Security starts with authentication, which verifies a person’s identity based on their credentials entered
at logon (usually a username and password). As an authenticator, the server is an authority in the network,
negating the concept of a peer-to-peer network in which all computers are equal. Once authenticated by
the server, the user gains general access to the network for which the server is responsible. Thereafter,
when the user attempts to access network resources, the authorized user account is compared against a
list of users or groups with access to those resources. If the user account is on the list, the NOS grant
access. This process of verifying that the user has the ability to access the resource is known as
authorization (see Exhibit 1-2).

18
 Exhibit 1-2: The server authenticates users and authorizes access to network resources

In a server environment, authentication better protects network resources than password protection in a
peer-to-peer network. Each user account is stored on the server only, mitigating the need to create user
accounts on each workstation. Recall that in a peer-to-peer environment, when a user wants to share a
folder, he or she must protect it with a password or have sufficient administrative rights to create user
accounts for those who want to access it. In a client-server environment, instead of using a password or
creating user accounts, permission to access the folder can be granted by the user (or an administrator)
to a list of users or groups of users on the server. The administrator creates the users and groups, and it
is not necessary to grant administrative rights to individual users. Exhibit 1-3 shows a list of users (shown
with a single-head icon) and groups (shown with a double head icon) on a Windows 2000 Server computer.

Exhibit 1-3: Granting access to users or groups

19
For added security, consider using emerging technologies such as smart cards to authenticate users. A
smart card includes an encrypted certificate and the user typically inserts the card into a reader attached
to the computer and enters a password or personal identification number (PIN). Because it requires both
the physical card and a password or PIN, the smart card protects against unauthorized persons using
stolen passwords. Similarly, a SecurID card synchronizes with the computer to generate a randomized
pattern of six-digit numbers displayed in a timed sequence; you enter what is displayed along with a PIN
as your password and you are granted access based on this “one-time” password. This “something you
have and something you know” authentication is similar to smart cards, but insertion into a reader is not
required.

Permissions
Authentication verifies the user’s identity, and authorization compares that identity against a list of users
or groups that are permitted to access resources such as files. Once a user accesses a resource, however,
there is the further question of what that user can do with that resource. That is what permissions are
about. Permissions specify the degree to which a user or group can access or alter a resource. For example,
an administrator could set permissions on a folder for individual users or for groups. Who should be able
to read but not change the folder contents? Who should be able to access the folder and change
permissions? It could be extremely problematic if permission were given to the wrong users or groups to
access sensitive information. Consider what would happen if everyone on a company network were able
to read the Salaries.xls spreadsheet! To avoid this kind of trouble, administrators assign specific
permissions to individuals or groups of individuals. For example, the administrator might assign
permissions to the Human Resources department that allows the ability to read and change the
Salaries.xls spreadsheet, assign read permissions to managers, and deny permissions to all other users.
The NOS permits the administrator to assign these very specific permissions for resources, as shown in
Exhibit 1-4.

20
 Exhibit 1-4: The NOS (Windows 2000 in this case) allows you to configure specific permissions
Of course, even on a peer-to-peer network, a user can set (perhaps unwise) permissions on a workstation
resource. However, because a server centralizes resources, it makes monitoring those resources more
efficient. Compare monitoring permissions individually on 10 computers in a peer-topeer network with
monitoring permissions for tens or hundreds of workstations on a single server.

Security boundaries
Closely associated with security, authentication, and permissions is the issue of security boundaries, which
delineate where a user can “go” in the network. In a large enterprise, administrators do not want an
employee in Los Angeles to be able to access a resource in Hong Kong only because they work for the
same company and are connected to the same network. To control user access, administrators can use
security boundaries such as the domain model of Windows NT 4.0. Users in the Los Angeles domain can
access resources only in the Los Angeles domain. Similarly, users in the Hong Kong domain cannot access
resources in the Los Angeles domain. However, the administrator of each respective domain can manually
allow access between the domains using what is known as a trust. Windows 2000 also uses a domain
system, except that it implements automatic trusts between domains—administrators do not have to
manually create a trust.

Performance
Because a server typically has significantly more powerful processors, more hard disks, and greater
storage capacity than a workstation, it can perform tasks on behalf of the workstation, improving overall
performance and freeing the workstation to perform other tasks.

For example, a special-effects technician creating a video clip can use the server to render the clip instead
of using his or her own local workstation. High-end servers that provide better performance can also cost
several times more than a low-end workstation. It is more financially feasible to purchase a $20,000 server
to service 10 special-effects technicians than it is to spend $3,000 to upgrade each technician’s individual
workstation (for a total of $30,000). In this simple example, the administrator saves the organization
$10,000 by using a server instead of upgrading individual workstations.

Processor
Three primary factors—clock speed, data bus, and cache—contribute to the effective speed of the
processor (CPU).
• Clock speed is the number of cycles the processor can execute in a single second, measured in
millions of cycles per second, or megahertz (MHz).
Instructions executed by the processor require a certain number of cycles, so the more cycles the
processor can handle per second, the faster it operates, or “thinks.” Current processors are also
superscalar—that is, they can execute more than one instruction in a single clock cycle.
• Data bus refers to the number of data bits that can pass into or out of the processor in a single
cycle. Data bus width is typically 32 bits. Think of a 32- lane highway over which data travels. Some
new processors offer 64-bit bandwidth, equivalent to a 64-lane data highway. Because of other
internal engineering modifications, a 64-bit processor utilizes its resources more efficiently than a
32-bit processor and is, therefore, more than twice as fast. Data bus width and RAM have a direct
relationship such that a 64-bit processor can also utilize 4,294,967,296 times more memory than
a 32-bit processor (assuming that the 64-bit architecture extends throughout the entire system).
• Cache is memory that exists on or near the processor itself but is separate from main system
memory (RAM). Processor cache stores recently accessed data from the hard disk or RAM. When

21
 the same information must be retrieved at a later time, the system can access instructions stored
in the cache more quickly than the system can retrieve information from the hard disk or RAM.
Also, the cache anticipates what the processor will request next and fetches it in advance, hoping
that it has guessed correctly. (Intel claims that 90 percent of the time, the cache guesses correctly.)

Although clock speed, data bus, and cache are components of both servers and workstations, servers are
typically more powerful in each respect. Also, servers often utilize symmetric multiprocessing (SMP),
which is the simultaneous use of multiple processors on the same server. SMP results in a corresponding
increase in performance such that two processors are about twice as fast as one; four processors are
about twice as fast as two, and so forth.

Hard Disk
Workstations and servers both store programs and data on hard disks. However, server hard disks are
often optimized for high performance and throughput—a measure of the quantity of data sent or received
in a second. For example, a 100 MBps throughput sends or receives at a maximum of 100 MB per second.
In most computers, the hard disk is the slowest component and, therefore, a bottleneck that inhibits the
overall performance of the system. Typically, the processor and memory can function thousands of times
more quickly than the hard disk and often must wait for the hard disk to read or write data before
performing other tasks. The following characteristics contribute to improving hard disk performance:
• Servers are capable of implementing multiple hard disks while addressing them as a single logical
disk. For example, drive E might appear to the NOS and users as a single drive. However, an
administrator can use special hardware or software to implement several disks at once that act as
a single drive E, thus reducing or eliminating a hard disk bottleneck. As a rough approximation, you
could utilize two hard disks to double the performance of the logical disk.
• A buffer (or read cache) on a hard disk is memory that functions similarly to cache on a processor.
When the hard disk retrieves data, it can store part of the data in a buffer. Later, the CPU can
request the same data and it will be retrieved from the buffer, which is many times faster than
mechanically retrieving it from the hard disk. Also, because data tends to be stored sequentially
on the disk, a buffer can automatically read the next sequential data on the hard disk in
anticipation that it might be needed soon after. An optimized workstation hard disk might have
between 512 KB and 2 MB of buffer, while a hard disk optimized for server use might have 4 MB.
• Access time (also called “seek time”) is the time it takes for the hard disk drive head, which reads
or writes data on the platter, to arrive at the location of the data. Access time depends upon the
spin rate of the hard disk. Most workstation hard disks spin at 5400 or 7200 revolutions per minute
(rpm). Server hard disks usually spin at 7200, 10,000, or 15,000 rpm or higher. The faster the disk
spins, the more quickly the required data can arrive under the disk drive head. Access times for
lower-end workstation hard disks are around 9 or 10 milliseconds (ms), and high-end server hard
disks can be as low as 4.5 ms.
• The disk drive interface is the hardware connecting the drive to the computer motherboard. The
interface is perhaps the most important factor in hard disk performance. At the low end, the
workstation interface is usually an Advanced Technology Attachment (ATA), considered
synonymous with the Enhanced IDE (EIDE) interface, of which there are several types. Generally,
workstation interfaces can read data from the hard disk at between 33 and 100 MBps and can
connect up to four hard disks (two disks per channel). Server interfaces such as Small Computer
Systems Interface (SCSI) can read data at around 100 MBps and support up to 15 hard disks.

In addition to the performance factors that distinguish server processors and hard disks, storage capacity
is significantly greater on servers than on workstations.
22
Storage capabilities
Network servers require more hard disk storage space than typical workstations in order to serve files and
applications. A file server must have enough disk space available to store files for all users. Because most
workstations have less storage space than a server, users could fill up their available local hard disk space
over time. By storing files on the file server, an administrator can more feasibly monitor available hard
disk space than he or she could by monitoring disk space on each workstation. Should low disk space occur
on the server, the administrator could use any of several methods to add hard disks on the fly without any
disruption in service? This is made possible by hot-plug or hot-swap capability, which means that you can
add or remove a device without first powering down the computer.

Access to applications
Network applications are server-based programs that run in memory and on the processor on behalf of
other servers or clients. Running applications on the server (referred to as back-end applications)
minimizes or eliminates the processing, storage, and memory requirements for each individual client.
Instead, the client computer runs a front-end application requiring much less overhead. (A front-end
application is an application running on the client that retrieves information processed by a back-end
application.) For example, it is much more effective to run larger databases on a high-powered server than
on client workstations. Even with multiple users running the database and performing queries at once,
the server can usually outperform the same tasks on a workstation, because each workstation does not
have to download the entire database across the network.

Let’s say an organization has a relatively small database of 20 MB stored on a file server. Without using an
application server, users at each workstation access the database file and run the database application
locally. This means that each user accessing the database must download the entire 20 MB database. This
is quite taxing on network bandwidth and requires users to wait for the download to complete. However,
with an application server, the back-end database application and file stay on the server. If each record is
1 KB in size, and a user runs a query against the database with 30 matches, the client downloads only a
30 KB result as requested by the client’s front-end database application (Exhibit 1-5).

Exhibit 1-5: Only the database query results download to the client

There are several reasons for the performance gains with an application server, including the use of
multiple processors, increased RAM, and fast hard disks on the server. Also, downloading only query
results instead of the entire database greatly reduces the burden on network bandwidth. Examples of
network applications and services include:
• Database applications: Programs such as Microsoft SQL 7 or SQL 2000 perform many of the
functions and provide the benefits mentioned in the previous paragraph. Primarily, application
databases benefit from the server in its role as a data warehouse (a storage location for extremely
large databases) and from the server’s processing power.
• E-mail services: Users can, to some degree, utilize Internet e-mail with little if any involvement
from the network administrator. However, an e-mail server allows the administrator to perform
management functions such as monitoring, virus scanning, forwarding, integration with directory
services, clustering, security management, failover (an alternate system that takes over for a failed
23
 system), redundancy, and so forth. Many e-mail servers go beyond e-mail services to provide
collaboration with users’ calendars and videoconferencing.
• Network management software: This varies greatly from one product to another. Administrators
of smaller networks might not require network management software, but larger networks
require some sort of management software to manage client software distribution and licensing,
monitor user activity, monitor and manage network traffic, backup and restore data, manage
Internet services, integrate with other server operating systems, and more. Network management
software is particularly useful because it can e-mail or page the administrator when there is a
problem with a server. Examples of network management software include Microsoft SMS,
Computer Associates Unicenter TNG, and IBM Tivoli.
• Remote access: Even when away from their desk or office, users require access to the LAN. When
a user is away from the local network, he or she is a remote user. Many remote users telecommute
from home or connect to the LAN from another office. Remote access makes this possible in two
primary ways. First, the user can use a modem to dial into the network. The remote access server
also has a modem dedicated to remote access purposes, and receives the user’s call.

Once the connection is complete, the user experiences the same network access as when locally present
and connected to the LAN (except that network connectivity is only as fast as the modem, usually between
14.4 and 53 Kbps depending on phone line conditions). Second, faster remote access connections such as
digital subscriber line (DSL) are replacing the modem-to-modem connection from client to server because
they are faster and always on.
However, traveling users generally continue to use a modem to connect (from a hotel room, for example).

• Virtual private networks: Users can also connect to the network over a virtual private network
(VPN), which is a highly secured network connection that makes eavesdropping from unauthorized
persons nearly impossible. This connection can take place over the first connection type, a modem.
Increasingly, however, users utilize an existing public Internet connection to establish the VPN
connection. For example, a telecommuter with a high-speed Internet connection to his or her
home office can establish a VPN connection to the corporate LAN and experience virtually the
same security as if locally connected to the LAN. Windows 2000 utilizes routing and Remote Access
Server (RAS) and Novell NetWare utilize Network Access Server (NAS) to provide remote access.

Centralized management
Any time the network grows beyond two or three dozen users, the administrator’s role can become more
reactive than proactive. That’s because he or she must continually go from one computer to the next,
troubleshooting applications, helping users find their files, and so forth. Although adding a server does
not eliminate these tasks, its centralized management capabilities can certainly contribute to reducing
overall administration.

Centralized management means that administrators can administer servers from one place instead of
traversing frantically from one end of the building or campus to the other fixing problems. With the proper
NOS options enabled or network management software installed on a server, an administrator can browse
a user’s hard disk from the administrator’s own computer to help them find the files they are looking for.
Or if a user is experiencing difficulty using a program, the administrator can view the user’s desktop
without leaving his or her own seat. Sharing applications from the server also reduces administration
because the administrator can install, troubleshoot, and upgrade applications from a single server location
rather than on each individual client. Managing applications on a single server reduces administration,
but it also provides a single point of failure. If the server fails or the application stops functioning correctly,
24
all network users are affected. Consider having a redundancy plan. For example, multiple servers hosting
the application can allow users to continue using the application should one of the servers fail. Clustering
is an example of fault tolerance, which allows for continued service despite failure of a server or
component.

Cost effectiveness
While you, as the network administrator, see obvious benefits for installing a server, it might not be up to
you. Much of the time, some combination of management, accountants, or both determine what they
think is best for the network strictly on the basis of how much something costs. So when you propose
adding a modest $8,000 to the budget for a server and applications for your LAN of 30 users, be prepared
for the possibility of a knee-jerk reaction and a resounding no. Increasingly, administrators must show the
money sense in the budgets they propose, and in fairness to an organization’s financial affairs, they
should. As you prepare to propose new server equipment, know that well-planned servers eventually
recoup their initial expense and continued operation in long-term cost savings and increased productivity.
The following sections illustrate the savings in several important areas.

Storage
Workstations in many organizations are several years old. Applications, operating systems, and data files
on the workstations grow larger, but the hard disks do not. It would be much more cost effective to install
centralized storage on a server for all the users. Although the price per megabyte on highperformance
server hard disks might be higher than that of a workstation, you can save time (and hence money) by
performing a single hard disk upgrade on the server instead of upgrading each workstation’s hard disk.
Upgrading individual user’s hard disks might also involve the sometimes tricky process of moving the
existing operating system, applications, and data to the new drive without disruption or loss of current
settings.

1.4 PROCESSORS AND MEMORY

As operating systems and applications become increasingly complex, workstations require more
processing power and RAM. Individual workstation upgrades quickly become cost prohibitive. However,
you can upgrade one or more servers and transfer the memory and processing burden to the server, as in
the case of a large database. Also, to avoid the costly operating system and hardware upgrade for client
computers, you can leave the existing operating system and hardware, and make the workstations thin
clients. A thin client (similar to a dumb terminal) receives its operating system environment, including
applications and data, from the server. The server does all the work, and the client sends the input from
the keyboard and monitor and receives the output to the monitor. For example, users could boot to
Microsoft Windows 3.11 at the workstation and connect to the server to run the Windows 2000 operating
system as the primary environment. Microsoft Windows NT Server 4.0, Terminal Server Edition, Microsoft
Windows 2000 Server, and Citrix Meta Frame provide thin client services. Novell’s ZEN works integrates
with and enhances thin client services to provide powerful administrative tools and to better manage the
user desktop.

Savings, efficiency, and productivity

Efficiency and productivity are difficult to quantify and vary from one organization to the next. Most
managers and accountants to whom administrators must defend their budgets already understand that
computers are generally cost effective. When proposing to add server equipment or software, you must
demonstrate that the addition saves money, increases efficiency and productivity, or both.

25
For example, what would it cost a corporation to lose its customer database to a competitor? A company
with an Internet connection exposes itself to unauthorized access to the world at large. Intruders who
succeed in invading your network might not do anything harmful, but administrators must plan for the
worst in case someone destroys, steals, or alters data. An administrator can add a firewall, which protects
the internal LAN from the public Internet and is placed between the LAN and the Internet. By protecting
the network, a firewall saves the company from potential financial and productivity losses. A firewall can
be either a hardware or software solution. In a server context, it is a server with special software. For
example, Novell offers Border Manager as its software firewall.

Classes of servers
Having established the need for a server in several contexts, you must also determine what type of server
is appropriate for your organization. A poorly chosen server might be more expensive and more powerful
than current and future needs require, or it might be woefully inadequate, necessitating the purchase of
more equipment at a less efficient cost factor. For example, if an administrator purchased a single PC
server capable of supporting two processors, and after installation discovers that it cannot meet the
processing demands placed upon it, he or she must purchase another system. In this case, it might have
been less expensive to purchase a single midrange server with support for four or more processors in the
first place. Although the Server+ certification focuses primarily on PC servers, a brief discussion of the
mainframe, minicomputer, and midrange computer classes provides you with a broader understanding of
other server types. In addition, it helps to know something about mainframes because administrators
often connect PC servers to them.

PC servers
Basically, a PC server uses Intel-compatible complex instruction set computing (CISC) processors, as
opposed to reduced instruction set computing (RISC) processors from vendors such as Sun Microsystems
or IBM. At the low end, a PC server can be a workstation-level computer in terms of its hardware. For
example, a small insurance company with 12 desktop PC workstations has a need for a file and logon
server. The company could install NOS on one of the existing PCs and suddenly the PC becomes a server,
and the network changes from a peer-to-peer LAN to a client-server network. The fact that the “server”
has only desktop PC-level hardware does not change the fact that it is a server, and in this small LAN, more
powerful hardware is unnecessary. On the high end, a PC server can involve significantly more powerful
hardware; therefore, it has a much larger case than a standard PC workstation. (For purposes of this book,
assume that the server is a high-end server unless stated otherwise.) The server case also provides much
more space for expansion and requires more components, such as extra power supplies.

Other characteristics of a high-end server include:

• Hot-pluggable PCI slots: Peripheral component interface (PCI) slots store various adapters such as
disk controllers, video cards, and network adapters. A desktop PC workstation would require you
to shut down the computer before adding or removing adapters. A server with hotpluggable PCI
slots allows you to add or remove adapters while the server is on. This is a very useful feature in
mission-critical operations where you cannot afford the extra minutes required to shut down and
restart the server. Some servers offer this functionality by allowing you to shut down individual
PCI slots without powering down the entire system.
• Hot-swappable hard disks: These allow you to quickly remove hard disks without powering down
the system. They are useful for quick emergencies that occur when it’s time to upgrade a hard disk
or replace a failed one.
• Power supply: Servers usually have two or three power supplies to power the additional
component slots and drives, and to provide redundancy. If one power supply fails, the system
26
 can usually continue to operate until you replace it with another hot-swappable power supply
(Exhibit 1-6).

Exhibit 1-6: Hot-swapping a power supply

• Cooling fans: Redundant hot-swappable cooling fans help ensure that the system does not
overheat if one of the fans fails, because the other fan(s) can continue operating until you hotswap
the failed fan for a new one. This is a critical element to maintaining server uptime. (Uptime is the
continued operation of the overall server or specific components such as the hard disk, depending
upon the context.) An overheated system can quickly burn out or reduce the life of the
components, especially the processor, which is the hottest component of the system.
• Expandability: Servers have more PCI slots and drive bays, and are modularly designed so that you
can easily add or remove the components.
• Heavy-duty chassis: The chassis is the metal frame to which the motherboard is attached and
which forms the case structure. A server chassis is much heavier than a workstation chassis and
often weighs about 75 lbs, including the case and components. It is heavier because the metal is
of a heavier gauge, and the case is often larger to accommodate more internal components, such
as more hard disks. After adding all internal equipment (drives, power supplies, and so on), the
server can weigh over 170 lbs!
• Rack mountable: For many servers, when you order the server you can specify a tower (upright,
free-standing case) or a rack configuration. The rack mount consolidates space because you can
stack several servers and other network equipment in the same floor space, offering good density.

1.5 MAINFRAME COMPUTERS

A mainframe is a large and extremely powerful computer. Nearly all computers in the 1960s were
mainframes and usually filled an entire room. IBM was one of the first mainframe pioneers and continues
to lead the mainframe market today with its S/390 and ZSeries 900 systems. The term “mainframe”
becomes less definitive as minicomputers become increasingly powerful and some mainframes become
smaller in size. However, there are several characteristics that continue to distinguish a mainframe:
• Size: It’s just plain big. Many of the newest mainframes boast a relatively small size compared to
their predecessors, and require about 20 square feet of floor space, and weigh over 2000 lbs.
• Cost: Mainframe prices are in the range of “if you have to ask how much, you can’t afford it.”
Mainframes are still in the range of at least several hundred thousand dollars.
• Processors: Mainframes nearly always use multiple processors—up to 32 for Intel-based machines.
Also, mainframes frequently use proprietary RISC-based processors that usually run a version of
UNIX. Although the processors in a mainframe might be comparable to a PC in terms of MHz, a PC
processor is busy delivering data to and from its peripherals. This is very time consuming in terms

27
 of clock cycles. The mainframe utilizes mechanisms to offload I/O and avoid the negative impact
on performance.
• Bus speed: No computer can be any faster than its main board (or motherboard) bus speed. While
a PC server might have a speed of 100 MHz, a mainframe might have significantly higher bus speed,
which when multiplied with its SMP capability can provide blazing processing performance.
• Memory: Mainframes can have upward of 96 GB of main memory. Also, mainframes have
extremely well-engineered cache and memory architecture, providing very high efficiency and
performance.
• Storage: Mainframe disk storage is capable of storing hundreds of GB or a terabyte (TB) or more.
Typically, large quantities of storage are external to the server.
• Durability: Mainframe hardware circuitry design detects and corrects errors, and it can anticipate
and alert administrators of impending problems. Some mainframes can generate a list of parts
that need replacement at the next regular maintenance. Mean time between failures (MTBF) is the
anticipated lifetime of a computer (or one of its components). Mainframes often have MTBF of
about 20 years. While it is difficult to specify PC server MTBF, it is usually a fraction of that of a
mainframe.

The minicomputer or midrange computer

Minicomputer characteristics lie somewhere between the desktop workstation and the Mainframe
computer. The minicomputer is sometimes known as the midrange computer. This is an extremely broad
range, with distinction between midrange and PC blurring at the low end, and between midrange and
mainframe at the high end. The first minicomputer was the Digital Equipment Corporation (DEC) PDP-1,
which cost $20,000 in 1959 and was not as large as a mainframe but was still quite sizable. Compared to
mainframes costing millions of dollars, the price tag was extremely attractive. However, the PDP series
was not considered successful until the mass produced PDP-8 in 1965, which was a tabletop-size unit.
Minicomputers also offer an advantage compared to mainframes in that they require significantly less
floor space and weigh less.

1.6 SERVER PLANNING

Determining the business purpose

The administrator must choose the most appropriate server solution for his or her business environment,
while also providing adequately for user demand and usage of the server. Sometimes, the administrator’s
choices involve other mitigating factors such as providing for interoperability between operating systems.
Server administration does not end after the server purchase. You must also place server equipment in
optimal locations for both accessibility and disaster prevention. Planning the placement should be
documented with accurate and up-to-date diagrams. Server locations must be environmentally fit so that
heat, dust, and humidity do not adversely affect server uptime. Finally, the administrator must arrange
for the best security available to protect both the hardware investment and the intellectual property that
the server data represents.

Asking the right questions

The network administrator must have a business sense of how a server plan directly benefits the
organization financially or indirectly in terms of improving efficiency or productivity. It is obviously the
responsibility of the network administrator to obtain necessary network equipment and servers.
However, going a step beyond this basic responsibility by contributing to the bottom line of the business
will also contribute to your value in the marketplace. Increasingly, organizations seek somebody who not

28
only “knows a lot about computers,” but also knows how to get the most value for every dollar spent in
servicing the network.

Although each organization differs in what benefits its business, the following questions provide a
common starting point as you consider if and how a server or servers can improve your organization’s
business and support its goals.

Do we really need this server right now?

As an administrator responsible for thousands (and sometimes millions) of dollars worth of equipment,
your IT budget can be substantial. However, you must take server purchases seriously and proceed
cautiously, just as if you had a minimal budget.

Although this book does not intend to force a financial philosophy, carefully weighing the need for a server
against its costs can sometimes help your standing as an administrator. For example, most publicly traded
companies watch the timing of quarterly expenses very carefully. If the company is experiencing less than
stellar performance in the quarter, it might not be the best time to propose a large equipment purchase,
especially if the purchase is not urgent (i.e., not particularly time-sensitive).

You might wait until the beginning of a more promising quarter and attempt to push through equipment
approval before anyone has a chance to get nervous! Conversely, there is little reason to pause before
purchasing a vital server. For example, if an e-mail server experiences a catastrophic failure, you would
want to replace it (or the failed component) immediately and without apology, because e-mail is a highly
utilized function of most organizations and used nearly every second of the workday. In this case, the e-
mail server purchase is urgent, and failure to purchase it would negatively impact your standing as an
administrator. In between the urgent and non-urgent equipment purchase comes the “we really could
use that pretty soon” purchase. This might be something that does not immediately affect daily
productivity or profitability, but for which you can provide reasonable justification. These types of
purchases might optimize an existing server plan or proactively save money in the long run. For example,
let’s say you have a growing research division that is vital to your company’s success. The research division
can get by on the equipment it currently has, but you often hear about slow access to large user files on
the department file server. The network has plenty of available bandwidth and acceptable network
utilization, so network issues do not seem to be a problem. (Bandwidth is the transmission capacity of the
network. For example, most Ethernet networks can transmit at 10 Mbps or 100 Mbps. Network utilization
is the percentage of bandwidth in use during a given period of time.) However, you have confirmed that
the file server is over utilized. Your solution is to purchase an additional file server to provide load
balancing by distributing the files between the file servers, effectively halving the burden on the original
file server. An additional benefit of load balancing is failover—if one of the servers fails, the remaining
server(s) continue to provide service. Although you might not be able to specify the exact long-term
savings, you can probably present an obvious case for improved productivity and redundancy with the
additional file server.

Is the expense of the server offset in savings?

Suppose that your research division also has an office in Denver. You find that researchers in Denver often
download large files from file servers in your local research division. The WAN link that connects these
two offices is a high, ongoing expense because subscription charges accrue according to usage. (A WAN
link is the telecommunications connection that links the various networks that comprise your WAN.) The
WAN link is oversubscribed (bandwidth utilization is so high that it slows network access), so you want to
reduce its usage. In this case, consider adding a file server to the Denver research office that periodically
29
synchronizes (updates two or more copies of the same file so that they are the same) with your location’s
research division. By controlling when this synchronization occurs, you can save bandwidth and lower
subscription costs. As an added benefit, the Denver researchers obtain faster access to their file resources.
In this case, you can easily justify the addition of the Denver file server against the savings of the
oversubscribed WAN link.

Is the timing of a server purchase appropriate?

Even if budgeting and financial resources are not a concern, the administrator must continue to be careful
about the timeliness of a purchase. For example, you might delay the purchase of a new server with the
fastest and latest processor even though you require more processing power. If you do not urgently need
that much speed now or in the near future, consider other possibilities. For example, if the existing server
supports SMP, and you have room for an additional processor; add another processor instead of
purchasing an entire server. You might also consider purchasing a server with a processor that is a step or
two below the latest and greatest—prices often come down after the release of a new processor.

Is sufficient expertise available to operate and maintain the server?

Sometimes an organization needs a server but does not have anyone available to run and maintain the
equipment. This is particularly true in the case of a WAN where multiple, geographically distant locations
make it impossible for a network administrator to be everywhere at once. This situation is difficult in that
you must supply the server needs of the organization, yet the server cannot run and maintain itself.
Although you will probably have to go ahead and provide the server, you might have to budget additional
funding to hire outside support services for that office or to fund travel expenses for you to physically visit
the server.

While software exists that allows you to remotely operate a server from your location, such software
cannot perform every function. For example, software cannot perform physical actions on the server, such
as replacing a power supply.

Anticipating user demand

It’s essential to remember that the network or systems administrator provides a service to users.
Remembering this responsibility affects the administrator’s perspective and job performance. Some
administrators find humor in belittling the end user. However, you will be a better administrator if you
take the user more seriously. For example, in asking for a faster Internet connection, users might not
understand that they are really asking for another high-speed Internet line and associated equipment
costing thousands of dollars per month. The request might or might not be valid. The administrator often
helps others (such as members of management) to determine which requests are valid.

The network administrator’s job is to ensure that users can perform their jobs in an efficient, timely
manner. To do this, he or she may need to wear many “hats,” from educator to technician to business
manager. The needs in each networking environment vary. However, the following questions might help
you as a starting point in servicing user needs.

How many users will connect to the server?

This is perhaps the most relevant issue in providing a server for user access. If you anticipate only a few
users, you could use virtually any kind of server for general purposes, such as a desktop PC server.
However, if you expect frequent access to the server from a large number of users, a more capable
midrange computer might be more appropriate with the fastest network card that your network design
supports. For example, a large corporation with a company intranet might have several midrange
30
computers serving as intranet Web servers. (Multiple Web servers providing the same Web content are
collectively referred to as a Web farm.)

A network interface card (NIC) is the computer’s adapter card that connects to the network and through
which network communication takes place. Consider installing two or more NICs in highly utilized servers.
This multiplies the effective network I/O to and from the server by the number of NICs you install. For
example, three NICs provide three times the network I/O capacity of a single NIC, provided the network
bandwidth is not oversubscribed.

What is the nature of the user access?

You might have only a few dozen users who, because of their access needs, require a more powerful server
than hundreds of users with a different type of access requirement. For example, a few dozen users
frequently performing complex and demanding queries on a very large database might require a server
with significant storage and processing power. However, hundreds of users requiring a simple logon server
probably do not need a significantly powerful server.

Should users be able to access the server directly?

Some servers are not designed for direct user access. For example, in a Transmission Control
Protocol/Internet Protocol (TCP/IP) network, a Dynamic Host Configuration Protocol (DHCP) server
assigns unique identification to each computer on the network. This function requires minimal processing,
memory, and hard disk storage, and a fairly bland server in terms of power can probably service hundreds
of computers. This type of server requires no direct access on the user’s part. As a rule, assume that users
do not require direct server access and take measures to prevent such access unless you know users or
groups of users who require a specific level of access. Also, user access to the server should only be from
workstations across the network; users should never log on locally to a server.

Planning for interoperability

Many network environments use multiple operating systems across their servers and clients (see Exhibit
1.7). Reasons for this mixture vary. Perhaps two companies with differing operating systems merge, or a
particular application only works with one NOS but you want to use the features of another NOS. The
administrator must ensure that different operating system platforms can operate with one another and
that the user experience is not disrupted. Ideally, the user has no idea that two or more operating systems
are in use.

31
 Exhibit 1.7: A mixed environment utilizes multiple operating systems

Sometimes an organization utilizes multiple operating systems because of a gradual evolution within the
company. Perhaps the organization started out with UNIX application servers and Windows 95
workstations, but later added Novell NetWare file and print servers for users to store and print files. (UNIX
usually operates on more expensive RISC-based processors. Linux is a version of UNIX and operates on
PCs in addition to Alpha RISC-based processors and PowerPC processors.) The expensive UNIX hardware
platform became less attractive compared to a less expensive PC computing platform, so PC servers
running Windows NT 4.0 and Linux were added. Later, the organization upgraded client computers to
Windows 2000 Professional and the UNIX hardware platform to PC servers running Linux. Regardless of
the path an organization takes to arrive at its current constellation of operating systems, the network
administrator must ensure that all operating systems interoperate as seamlessly as possible.

Operating system vendors have come to realize that no one operating system will meet every need, and
they have produced various patches and other software to allow interoperability. For example, a Windows
NT 4.0 logon server and a NetWare file and print server may have to interoperate. In this case, you
probably want users to log on to the Windows NT 4.0 server as usual and access files on the NetWare
server. Users should not be required to log on to the NetWare server separately if you want seamless
interoperability. Instead, users access resources on the file and print server without any awareness that it
is a NetWare server. Microsoft includes Gateway Services for NetWare with Windows NT 4.0 (see Exhibit
1.8). This product allows users to access NetWare resources through the Windows NT 4.0 computer.
Whatever network environment you find yourself in, research the various operating system combinations
and find the resources that allow the best interoperability.

32
 Exhibit 1.8: Gateway Services for NetWare allows access to the NetWare server

Consider protocol incompatibilities in the mixed environment as well. Continuing the previous
interoperability scenario involving Windows NT 4.0 and NetWare, realize that Windows NT 4.0 installs
TCP/IP by default, and NetWare often uses IPX/SPX (Internet Packet eXchange/Sequenced Packet
eXchange). These two protocols are incompatible.

For interoperability purposes, you would probably install both TCP/IP and IPX/SPX on the Windows NT 4.0
server. Then, the Windows NT 4.0 server can internally translate from IPX/SPX to TCP/IP (and vice versa)
as messages travel to and from the NetWare server (Exhibit 1.8).

Exhibit 1.8: Using TCP/IP and IPX/SPX, Windows NT 4.0 can communicate in both protocols

Prior to adding another operating system to your network design, be sure to test for possible
incompatibilities. For example, perhaps your UNIX system uses a custom designed database. If you plan
to install Windows 2000 Server, it will not be able to use the same application. You will either choose a
different database product or pay to port the application to the Windows platform.

Also consider the interoperability of network operating systems with client workstation operating
systems. For example, if you install a NetWare server in a LAN that uses Windows 98 computers, can the
Windows 98 computers access all the benefits of the NetWare server? In fact, they cannot do so using the
Microsoft Client for NetWare Networks software that is included with Windows 98. In order to utilize the
powerful features of Novell Directory Services (NDS), Windows 98 clients must download and install the
Novell Client for Windows 95/98 from Novell. (Novell Directory Services [NDS] is an example of a directory
service, a hierarchical database of network resources that allows users from anywhere in the enterprise
to access resources throughout the organization. Microsoft Active Directory is another example of a
directory service.) An enterprise is a geographically dispersed network under the jurisdiction of one
organization. It often includes several different types of networks and computer systems from different
vendors. If you decide to convert from one NOS to another, be aware that the user and group accounts
are incompatible between operating systems. For example, if you want to convert from NetWare to

33
Windows NT, the NetWare accounts are not readable by the Windows NT operating system. However,
most NOS vendors offer a migration tool that allows you to separately convert the accounts.

1.7 SERVER PLACEMENT

No matter how good your business sense, ability to meet user demands, or skills in server interoperability,
a service outage and lengthy recovery significantly impact the business of the organization.

One of the most critical issues in server planning is assessing the physical location at which you plan to
install the servers. An improper server environment can result in severe problems later on. Installing a
server involves much more than finding an empty space, plugging it in, and installing the operating system.
You must also place the server so that it serves network users in the best possible way. In a global
enterprise, also consider factors such as site links and bandwidth utilization within and between networks.

For example, you might be from the United States and accustomed to reliable, high speed WAN links.
However, many locales have slower or less reliable connections. A WAN link connects a site, which consist
of the LAN(s) on either side of a WAN connection. Exhibit 1.9 illustrates a WAN with two sites and two site
links.

Exhibit 1.9: Two sites utilize a WAN link and a modem link

If the WAN link is slow but is the fastest type available in that locale, consider adding an additional WAN
link of the same type. For example, the best available connection at some locations might be a 56 Kbps
ISDN (Integrated Services Digital Network) from the telephone company. The advantage of an ISDN
connection is that you can either leave it in an always-on state or have it dial out to another site on
demand (which is useful if the Telco charges for connection time). Also, an ISDN line usually provides
clean, consistent data transmission. If this single connection is too slow, you might be able to combine
two ISDN channels and double the speed, or add a dial-up modem connection to increase bandwidth
during peak times (again refer to Exhibit 1.9). The dial-up connection can also provide redundancy in case
the ISDN line fails. Server placement within your WAN has n impact on network service and response to
user requests. First, let’s look at server placement within a site, and then at server placement for
connecting sites.

1.8 INTRA-SITE SERVER PLACEMENT

Intra-site communication refers to communication between hosts within a single site, often over a LAN.
Within the site, you should determine the best location in the building and on the network to place the

34
server. Even a high-powered server’s performance can flounder if the server is poorly placed. Although
most network connections in a LAN are high-speed and well-connected, you should still carefully place
the server in the most efficient location possible. Most networks represent a progression of growth both
technologically and physically over a period of years. As a result, a network might have some locations
that offer better connectivity than others. Exhibit 1.10 shows a network that originally started with a 4
Mbps network. As the network grew, a new section utilized a 10 Mbps network. Later, a 100 Mbps section
was added. However, the servers remain in the 4 Mbps section.

Exhibit 1.10: A network with various bandwidth speeds

Users on the 4 Mbps network seldom complain of bandwidth speed connecting to the servers because
they have never experienced speeds faster than 4 Mbps. Users on the 10 Mbps network complain more
often, and users on the 100 Mbps network complain every day about slow service from the servers
because those users can compare 10 or 100 Mbps speed to 4 Mbps speed. The connection speed problem
compounds with outside sources accessing the servers. For example, if one of the 4 Mbps servers were a
File Transfer Protocol (FTP) server, the effectiveness of the server file transfer would be severely limited.
It would be better to move the servers from the 4 Mbps network to a faster section, or to redesign the
network so that all sections are at 100 Mbps. Though these seem like simple solutions, in some
environments making these changes could be a difficult task for several reasons. Perhaps the 4 Mbps
network is large, and it requires a significant equipment investment to upgrade all equipment, including
hubs and the NICs, on all servers and clients. Even if it were impractical to upgrade the 4 Mbps network,
you would still want to move the servers. Moving a server might also prove to be problematic if the higher-
speed networks do not have the physical facilities necessary for such a move—such as if a server room or
closet were not available. If servers were added to the 10 or 100 Mbps networks that replicate data with
servers on the 4 Mbps network, then the 4 Mbps network would experience even higher bandwidth
utilization, limiting the effective shared connection speed for all other network hosts.

35
Inter-site server placement
Inter-site communication refers to communication between hosts in different sites, often over a WAN link.
Servers communicating over a WAN link also require you to determine the best connection method to
optimize your bandwidth. When planning inter-site network communication across a WAN, you should
generally place the servers that directly communicate across the WAN in closest physical and logical
proximity to the WAN link. This makes sense because it reduces the number of variables between the
WAN link and the server. In addition, placing the server as close as possible to the WAN link helps ensure
that the server receives the maximum available bandwidth from the WAN.

If you place the server further away from the WAN link, you increase the number of possible problems.
For example, additional hubs, switches, or routers (network devices that divide the network into separate
parts usually known as subnets and forward network traffic to appropriate destinations) each add a
potential point of failure in the data path between the WAN and the server. For example, in Exhibit 1.11
a Windows NT 4.0 domain controller is moved closer to the actual WAN connection, removing potential
intermediate points of failure. Also, placing the server further away from the WAN link slows down
network performance, because data traveling from the WAN must compete with other network traffic
that occurs within the LAN.

Exhibit 1.11: Inter-site server placement can improve availability and performance

1.9 CREATING THE NETWORK DIAGRAM

36
A network diagram is a physical and/or logical representation of the network; it is also known as a network
map. You create a network diagram to design a network, keep a record of the network, or assist in
changing or troubleshooting a network. You can draw a network map by hand, but regardless of your
artistic skills, you should use special software designed to create a network diagram (see Exhibit 1.12). As
a technician, you will find it much easier to trace problems if you have a diagram of what the network
looks like. For example, if you visit a large network and you have little knowledge of its physical or
topological structure, attempting to physically locate a failed server might be difficult without a network
diagram.

The network diagram also provides a record and justification for why an organization decided to use
specific servers and why the servers are placed in their specific location. The diagram further helps to
logically determine a course of action in making a change to the network design and in troubleshooting.

Exhibit 1.12: Network diagram (Visio)

1.10 PLANNING PHYSICAL SITE READINESS

The physical server environment is one of the most critical aspects in determining where to place servers.
In fact, you might have to significantly modify a particular room to ensure that the conditions are optimal
for server reliability and uptime. Variances in the physical environment can also affect the lifetime of the
server and its components. The two most significant factors affecting the health of a server are
temperature and humidity. Other elements about the site’s physical readiness include floor space, power
availability, and the possibility of a fire or flood. The physical site plan for your server room is vital to the
success of your network. However, you cannot plan a new server room yourself, regardless of how much
you know about servers and networking. Several key planning considerations require the involvement of
an architect, electrical and mechanical engineers, and a general contractor. For example, you probably do
not want to design a fire suppression system on your own. For that, you would use a mechanical engineer
to ensure the best possible solution for your environment and to avoid liability on your part and that of
your organization. Also, these professionals can ensure that the server room installation complies with
federal OSHA requirements and local building codes.
37
Temperature
Servers run hot. When you consider individual factors that contribute to server heating and put them all
into a single box, temperature problems quickly compound. The hottest element is and will probably
always be the processor(s). The processor consists of around 40 million tiny transistors, each charged with
electricity, albeit a small amount at about 1.7–3.5 volts. (A transistor is an electronic device that opens or
closes, or turns on or off to provide a logic gate or switch, and provides the “thinking” capability of the
processor.) Maximum temperature tolerance for the processor is about 185° Fahrenheit (85° Celsius);
however, you should never allow the temperature to get this high. Fans and other cooling measures
dedicated to the processor help to keep the temperature at 90–110° F (32–44° C). If you have an SMP
system with multiple processors, potential temperature problems multiply accordingly. Other hot
components in the system are the hard drives, which on a server often have cooling fans of their own,
and the motherboard. Internal components in the server collectively contribute to the overall heat of the
server room. The increased heat in the air results in warmer air entering the server and aggravates the
heat issue, which is why the server room should have dedicated air conditioning. Generally, a series of
fans inside the case help keep the system cool and achieve maximum effectiveness if the ambient (i.e.,
surrounding) room temperature is also cool. In order to provide cool ambient temperature, keep the air
conditioning in the server room as cool as possible. To compromise between human comfort and cooling
server equipment, you can usually keep the server room temperature between 68 and 72° F (20–22° C).
However, be sure to keep the temperature at a constant setting, because temperature fluctuations cause
expansion and contraction of server components, shortening the server’s life span. Because equipment in
the server room generates heat disproportionate to the heat level in the rest of the office, setting a
thermostat in general areas cannot adequately cool the server room. A thermostat set at 70° F (21° C) for
general areas will allow significantly higher temperatures in the server room. Therefore, it is important to
provide an independent air-conditioning system and thermostat in the server room. If the budget allows,
also consider installing two air-conditioning units and thermostats in the same server space for
redundancy (Exhibit 1.13). If one unit fails, the other should be able to cool the room.

Exhibit 1.13: Use separate thermostats and air conditioners in addition to internal equipment fans to cool equipment

Regardless of the ambient room temperature and adequacy of server cooling fans, the server cannot
adequately cool itself unless you provide good airflow to the server. This is an often overlooked factor for
a number of reasons. For example, server rooms often lack adequate space. As administrators cram more
equipment into the server room, they are often forced to shove equipment closer to the wall to create
more usable floor space Or, someone receives a new carton of equipment and, for lack of a better place,
38
simply places it in front of the server rack—unaware that he or she has just blocked the ventilation slots
to the server. Space in front of and behind the server or server rack is critical. According to most
specifications, about three feet of clearance is required both in front of and behind the rack (Exhibit 1.14).

Exhibit 1.14: Allow three feet in front of and behind server equipment

Air quality
While air quality sounds like a health-related issue for people, it is also an important issue for server
equipment. Air quality in a server context means that the air must be clean and as free as possible from
dust particles. Rooms designed to house server equipment often have added filtration beyond the usual
air filters installed in the general-purpose HVAC (heating, ventilation, and air conditioning) system.
Excessive dust in the air directly relates to the previously discussed issue of temperature. A layer of dust
effectively becomes insulation on server components. Insulation is fine for a home, where you want to
keep heat inside, but it’s not fine on server components, where you want to prevent heat as much as
possible. Also, dust particles can adversely affect moving parts such as floppy and hard disks. Dust
accumulation can also present a fire hazard. Take regular measures to clean dust off of all server
components, regardless of air cleanliness. Add supplemental air filtration to any server room that does
not already have it. Remember that dust begets more dust. Passersby or air movement from the HVAC
system easily disturbs a layer of dust. While you as the administrator are responsible for maintaining dust-
free components, a cleaning service can remove dust in general areas. There are contractors who
specialize in cleaning controlled environments, such as your server room, using highly trained crews. An
example is Data Clean Corporation. Several server components—such as the power supply fans,
supplemental cooling fans, server racks and cabinets, and hard disks—might also include supplemental
filters. For example, a force-filtered server uses one or more filtered fans to supply main internal airflow
throughout the server. Other cooling fans inside the server only draw upon this filtered air, creating a
positive pressure environment (Exhibit 1.15). In server rooms that are extremely sensitive to dust, you can
also install an adhesive pad in front of the server room door that collects dust particles from the bottom
of shoes as people enter the room.

39
 Exhibit 1.15: A positive pressure environment helps ensure clean air inside the server

Humidity
Humidity factors vary widely depending upon the physical location of the site. For example, the desert
climate of Phoenix is not as likely to present the same humidity issues as the rain-soaked climate of Seattle.
Humidity affects the health of electronic server components because a drier environment presents a
greater occurrence of electrostatic discharge. Electrostatic discharge (ESD) is static electricity that can
damage, destroy, or shorten the life of the server’s electrical components. Many servers specify operating
allowances between 20 percent and 80 percent non-condensing relative humidity (non condensing means
there is no moisture accumulation, such as on the outside of a cold glass). However, you should strive to
humidify or dehumidify the air as needed to keep the humidity range between 40 percent and 60 percent.

High humidity presents the possible problem of condensation on equipment, which could obviously drip
onto electronic components and generate significant damage—not to mention an electrocution and fire
hazard to personnel. Even high-humidity environments do not normally cause condensation unless the
temperature changes drastically—perhaps due to an HVAC outage. Higher humidity can create corrosion
on metal components such as adapter cards and memory chips, and accelerate deterioration of magnetic
media such as tape backups and floppy disks. Few environments contend with high humidity because both
heating and air conditioning automatically remove humidity. However, you might find higher levels of
humidity in basements or other subterranean locations, or environments that do not have a quality HVAC
system in place. If humidity is too high or too low, HVAC companies can offer a variety of solutions to add
or remove humidity. Typically, HVAC modifications add humidity to the air that flows from the heating or
air-conditioning unit. Utilize a dehumidification solution from the HVAC Company as well. For economic
reasons, you might be tempted to utilize a household dehumidifier in smaller server rooms. However,
these units cannot be recommended for at least two reasons. First, these units are designed to cycle on
and start dehumidifying based on a humidity threshold setting. When the unit powers on, it might cause
a brief dip in power level if it is on the same circuit as server equipment. Second, these units usually
remove the humidity from the air by condensing it into water in a pan. The pan must be emptied
regularly—an overflowing pan threatens safety (slick floors) in addition to electrical hazards.

40
Flooring
In a server room, flooring is much more than a location on which to place equipment. Flooring can have a
direct effect on the health of your servers, particularly in respect to the risk of ESD and the efficiency of
cooling. In practical terms, flooring also affects where you put cable and smoke alarms. Choose from either
a flat floor or a raised floor—each has its own characteristics and advantages.

A flat floor usually involves commercial-grade floor tiles on top of concrete. Check with your architect for
floor tiles that can withstand the pressure of heavy server equipment and are static-resistant. Inevitably,
some equipment will scar or crack tiles, which is not a functional issue if the damage is only cosmetic. To
plan for future equipment additions and rearrangements in the server room, be sure to request extra
replacement tile. Also consider no-wax flooring to avoid the time-consuming and messy job of stripping
and applying wax, in addition to the regular maintenance of machine buffing, which can generate a flurry
of dust. Avoid carpet because it can retain dust and presents a static risk despite manufacturers’ best
efforts to make static-resistant carpet.

A flat floor requires you to place cable, power lines, and HVAC ducts inside walls and above ceiling tiles
between the actual ceiling and the dropped ceiling—a space known as the plenum (Exhibit 1.16). Running
cable in the plenum might not always be possible (for example, in an older building that does not have a
plenum area).

Exhibit 1.16: The plenum is useful for cable runs, HVAC vents, and power lines

A raised floor attaches to supports that provide a subfloor between the concrete floor and the floor panels
(see Exhibit 1.17). This space (also called a plenum or under-floor cavity) serves the same purpose as the
plenum in the ceiling—you can run cables, power lines, and HVAC vents in this space.

Exhibit 1.17: A raised floor rests on posts and provides plenum space

41
The depth of the subfloor varies from one design to another, but it usually involves 2-foot-by-2-foot panels
11 inches above the concrete. (You can adjust the height using different sized supports.) Raised floor
installations were quite common in the era of large mainframes, and then became less common as more
compact PC-based servers played a larger role. Server rooms became smaller as organizations
decentralized from a very large room containing all servers to multiple smaller server rooms. Now, many
organizations use a datacenter to house space-consuming equipment or consolidate contents of
departmental server closets into a single, centralized, larger server room. (Datacenter is a term with two
meanings, depending upon the context. It can refer to a consolidation of the majority of computer systems
and data into a main location, or it can refer to one or more very powerful servers optimized as database
servers— sometimes configured with as many as 32 processors. This context references the
former.)Heavy-duty floor panels are designed to withstand an enormous amount of weight (over 1000 lbs
each) and are often steel filled. Some floors use an I-beam construction for added stability at the edges of
the panel. These panels can be very heavy, weighing up to 45 lbs each, and require a floor puller with
suction cups to lift them off the supports (Exhibit 1.18). The floor surface is specially coated to reduce ESD.

Exhibit 1.18: A floor puller uses suction cups to remove floor panels

When running cable in either the ceiling or floor plenum, consider using cable trays or other cable
organizers to prevent a tangled mess and minimize future troubleshooting efforts. Also consider pulling
the cable through a conduit to minimize interference with other mechanical equipment and to make
replacing cable more feasible should the need arise. Raised floors (Exhibit 1.19) offer excellent grounding
to avoid ESD. Many designs ground each supporting post and offer grounding points for server racks,
cabinets, or other equipment. You can place HVAC vents beneath cabinets or racks to force cooled air up
through the equipment. A cabinet usually includes at least one 10-inch (25.5-cm) fan (and several
supplemental fans) at the top to draw cool air through the opening at the bottom and expel warm air at
the top.

Exhibit 1.19: Raised floors can assist in the cooling of server cabinets and racks

42
Coordinate raised floor installations with your architects, engineers, and general contractors. Raised floor
materials are available from many vendors. For a good start on the physical installation of raised floors
and server rooms in general, visit the following Web sites:
• www.beanfield.com
• www.accessfloorsystems.com
• www.compucraftconstruction.com

Power
Explanation Designing an appropriate power solution for the server room is one of the trickiest
considerations and should fall almost completely to the electrical engineers. Just tell the engineers what
you want and the level of redundancy you require, and depend upon them to provide the solution. When
providing planning objectives to the engineers, be sure to consider the factors of availability, quality, and
susceptibility to electromagnetic interference.

Availability
Three power sources provide power to the server room: the main power supply, the uninterruptible
power supply (UPS) for temporary power, and backup generator power for extended, system-wide
outages. Make sure to request dedicated circuits to the server room that are separate from the building’s
main power supply. Calculate the total power the current equipment requires, including servers,
monitors, routers, hubs, and switches. Add to this calculation anticipated amperage requirements for
future expansion, and ask the electrical engineers to oversupply the power requirements just to be sure.
(Also make sure to install plenty of easy-to-reach electrical outlets.) Although you should rely on an
electrical engineer to design server room power, you might also want to keep on hand the IEEE (Institute
of Electrical and Electronic Engineers) publication IEEE Recommended Practice for Powering and
Grounding Sensitive Electronic Equipment (ISBN 0-7381-1660-2). This book is also known as the “emerald
book” in reference to its cover color, and provides complex electrical information such as reducing
electrical noise and ensuring proper grounding. You might find it hard to locate in retail stores, but you
can obtain it from www.ieee.org. Administrators can probably determine what kind of UPS to use;
however, the engineer should plan main power and backup generator power. The purpose of the UPS and
backup generator is to provide redundancy, ensuring that power is available at all times. The UPS supplies
power temporarily while administrators perform a graceful shutdown of server equipment. Otherwise,
the sudden loss of power to the server can be extremely damaging to the operating system, applications,
and open data files. Exhibit 1.20 shows a typical power configuration for a server room.

Exhibit 1.20: Main power, UPS systems, and backup generator provide power to the Server room
Backup generators can be extremely expensive depending upon the amount of power they provide. Some
generators cost about $250,000 and require a facility of their own, separate from the main building.
43
Backup generators operate for as long as diesel fuel or natural gas is available. You will still need UPS
systems because it usually takes about 15 seconds for the generator to start up and supply power. Of
course, only organizations that absolutely require 24/7 operation would opt for such an expensive
generator, which might also provide main power to the rest of the organization.

Quality
Clean power extends the life of the server and its components. “Clean power” means the absence of
surges, spikes, dips, or poor grounding, which can lead to short circuits, tripped electrical breakers, and
possibly damage to equipment or people. Use a receptacle tester (also known as a polarity tester) to test
receptacles for power and grounding, especially in older buildings that might have questionable electrical
wiring. Request that the electrical engineer provide a connection to the earth ground in the server room.
The earth ground of a home typically connects to the plumbing outside that provides water into the home.
In a commercial building, electrical engineers usually design the earth ground to utilize a rod that is driven
deep into the ground. The electrical engineer takes measures to ensure clean power, but the
administrator also ensures clean power to the server by using good surge protectors, UPS systems, and
possibly line conditioners, which supplement power in the event of a brownout and/or minimize
electromagnetic interference (EMI).

Electromagnetic interference
EMI is a byproduct of electricity and can disrupt or corrupt data traveling along network cable as well as
disrupt other electrical equipment. Design data cable routes to and from the server room so that they
avoid electrical equipment such as fluorescent lights, heavy electrical equipment, motors, and so forth.
Make sure your electrical engineer is aware of the types of other equipment (such as heavy manufacturing
equipment) in your organization so that he or she can design around potential EMI pitfalls. Shielded
twisted-pair (STP) network cable includes a foil inner jacket, adding a level of protection against EMI
(Exhibit 1.21); however, you should avoid EMI sources when possible.

Exhibit 1.21: STP cable includes a foil inner jacket to protect against EMI

Because servers and associated equipment are electrical devices, they also produce a level of EMI. While
most equipment manufacturers take precautions to minimize the production of EMI, a certain level is
unavoidable. If you find that certain server equipment exhibits strange, intermittent problems that do not
seem to be associated with a specific component or the NOS, try moving the equipment to a different
location where there might be less EMI from surrounding equipment. Also be sure to cover any open drive
bays and expansion slots, and leave covers attached to servers when you are not servicing them.
Otherwise, these exposures can radiate EMI. If a situation arises in which you need to shut off a breaker,
the description next to the breakers might be blank, incomplete, vague, or just plain wrong. Verify that
you are about to switch off the correct breaker so that you do not inadvertently shut down other systems.
You can use a circuit breaker finder, which is actually two pieces. The first piece (the transmitter) plugs
into an electrical outlet on the circuit you wish to shut off. The second piece (the receiver) emits a tone
when you physically pass it over the correct breaker.

44
Disaster planning
Disaster planning requires a significant budgetary outlay for the relatively unlikely possibility that a natural
disaster, building defect, or other unexpected occurrence takes place. However, in the event that a
disaster does occur, you will be glad for every penny spent in planning. Much of this course teaches
disaster planning in various respects, such as the typical hard disk failure; however, this chapter focuses
on planning the server room to be as resilient as possible in case of fire or flood.

Fire detection and suppression

A fire presents an obvious threat to people and equipment. Most buildings are built (or remodeled)
according to relatively stringent OSHA regulations and local building codes. As a result, people sometimes
dismiss the threat of a fire. However, alert administrators must remain keenly aware of the possibility of
a fire (particularly an electrical fire) in both planning and daily operations. If there is a single location at
which a fire is most likely, it is probably a single room filled with complex electrical equipment—the server
room. First, in the event of a developing fire, you must ensure that people in the server room are notified
of the danger. Smoke alarms serve this purpose, and commercially available alarms can alert emergency
services and people within the organization in addition to setting off an audible alarm. While most smoke
detectors are in the ceiling, they can also be in the subfloor in a raised floor design. If they are in the
subfloor, you should mark the surface of the floor panel above the smoke detector with adhesive signs,
so that if it goes off you can more easily find it when it’s time to reset or maintain it. In years past, halon
was the primary fire suppressant in server rooms because water has an obvious negative impact on
electronics. Essentially, a fire is a chemical chain reaction. Halon breaks the chain reaction by substituting
hydrogen atoms with halogen atoms. Halon leaves little or no residue and does not cause electrical short
circuits. However, some suspect that halon can cause corrosion on computer components. Because halon
is harmful to the earth’s ozone layer, the Environmental Protection Agency (EPA) has banned the
production of new halon, although you can purchase recycled halon. Because halon is being phased out,
you might want to choose a different chemical fire suppressant. Consult EPA’s list of halon substitutes at
www.epa.gov/ozone/title6/snap/halonreps.html. Unfortunately, converting to another fire-protection
chemical involves more than replacing the gas canisters—you must replace the entire system. Whatever
product you use to extinguish fire, consider that you might also need an emergency ventilation system to
expel gasses from the suppressing chemicals or burnt objects.

If you choose to supplement the chemical extinguisher, consider using a dry system instead of a water
sprinkling system. A dry system uses water but only fills the pipes when there is a fire, ensuring that a
damaged or leaking sprinkler head does not harm equipment. When a fire alarm trips, the pipes fill with
water. Then, heat sensors release water from only areas of the room that indicate heat caused by fire.
This prevents the unnecessary release of water in areas where there is no fire threat. Also, when the fire
is extinguished, the sprinkler head can automatically close to prevent excess water release.

Consider including floor drains for quick removal of water. Make sure the drains include a backflow
prevention system so that sewer water backups do not flood the server room.

Flood considerations
In a flood, there is good news and bad news. The bad news is that there is not much you can do to prevent
a flood. The good news is, at least you have less planning to do. Depending upon the cause, you might be
able to prevent some types of flooding. For example, you could avoid placing the server room in locations
where plumbing runs above or below the floor. A burst pipe could cause immense damage to server
equipment. Also, remember to request floor drains with a backflow prevention system to prevent sewer
45
backups into the server room while allowing fire sprinklers or other sources of water to evacuate. If the
flood is caused by a natural disaster, placing the server room as centrally as possible and away from
exterior walls might allow other rooms to absorb the brunt of the initial water flow, although it may
eventually reach the server room. General plumbing principles dictate that your floor drains seldom work
when a large-scale flood is in progress. Your flood disaster plan will primarily determine the best way to
minimize damage and move equipment quickly, instead of trying to avoid the flood. In an evacuation plan,
move anything storing data first—file servers and backup tapes in particular. A wellplanned system design
calls for redundant copies of data, so you probably have off-site copies of your data. However, off-site
copies are usually not as recent as local copies. Even though the servers are expensive, they are
replaceable— whereas the organization’s data is probably not replaceable. Remember that much server
equipment is extremely heavy and cannot be lifted by a single individual. Also, some servers NOSs feature
built-in redundancy. For example, the Windows 2000 Active Directory automatically replicates its
database between domain controllers.

Using space effectively

In addition to technical issues relating to servers and networking, other factors can affect your server
planning. Some factors might be specific to your organization, and others might be common to most
situations, as in the following list:
• Choose a central location: If you have a choice (as in new construction), try to locate the server
room centrally (Exhibit 1.22). A central location makes it easier to provide good connectivity
to the rest of the network and saves cabling costs.

Try to design Internet and WAN traffic to and from your organization so that it flows freely and with as
few intermediary devices as possible (switches or routers, for example).

Exhibit 1.22: Centralize the server room for accessibility and reduced cable costs
• Consolidate space: The best way to get the most bangs for the buck out of server floor space
is to use racks and cabinets. (A cabinet is similar to a rack, except that it has a locked enclosure
46
 and cooling fans.) That way, you can store several pieces of equipment in a single horizontal
floor space.
• Restrict foot traffic: Place the server room in a low-traffic area to minimize the risk of
unauthorized access.
• Avoid exterior windows and walls: Place the server room away from exterior building windows
and walls for environmental and security purposes. In event of inclement weather, exterior
windows might leak wind or rain, and sunlight contributes unwelcome warmth to the server
room. A window that displays your server equipment to the outside world presents a security
risk because someone can surmise much about your network design by viewing the
equipment. Also consider that an evildoer could break the window and access the server room.
• Be prepared to budget extra financing for server room design: A server room is expensive. It
requires expensive equipment and special design considerations. If you are responsible for
budgeting the server room, bear this in mind and warn management in advance. As a general
rule, you can multiply the cost per square foot of the building’s general office space times four
to arrive at the cost of server room space.

Planning a secure location

Administrators of larger corporations are responsible for hundreds of millions and even billions of dollars
worth of vital company data, such as customer databases, company secrets, accounting records,
confidential employee records, and scores of other types of information—not to mention the value of the
server equipment. Even if the organization is small, protecting company assets is still the administrator’s
vital concern. Securing the server room protects both the physical assets and the information stored on
the servers. Although electronic security is designed to protect against hackers, viruses, malicious Internet
users, and the like, physical security protects the actual server facility and equipment.

Physical security can be sophisticated, but it is not complicated. It is as simple as physical security has
always been for everything from precious documents to currency—places the valuables in a safe and locks
it. The server room is really a very large safe, and you as an administrator are responsible for ensuring
that it is not accessed by unauthorized people. In fact, some corporate contractors to the United States
government run servers and a few highly secure workstations in steel vaults.

All other security measures are only as effective as the level of physical security.

You might have extremely restrictive permissions and detailed auditing records on who can access
company payroll records. While hackers are always a threat over the Internet or internal network, anyone
with physical access to the server can utilize special utilities that allow access to the data stored on the
hard disk even though they are not authorized for such access. For example, a utility known as NTFSDOS
allows the user to boot the server from an MS-DOS floppy and gain complete control of the NT file system
on an otherwise secure hard disk.

Monitoring security starts with your own IT staff.

47
Cutting your losses
A common source of company loss is employees, especially in the server room where there are a great
many valuables. Sometimes, a security breach could be an innocent oversight—perhaps a new
administrator provided a tour of the new server room to unauthorize outside persons. Or the security
breach could be a deliberate, mean spirited attack. Perhaps a disgruntled IT employee might want to seek
revenge against the company before leaving for another job and, using his or her administrative privileges,
destroys company data. It is wise to assume that anybody can steal or damage company assets—so
implement sound security measures to remove or reduce such opportunities.

Always log off before leaving your computer. Otherwise, any passerby can access the local computer and
network with the same authority you have, and any trace of improper activity will be logged to your
account. Operating systems such as Windows NT or Windows 2000 allow you to lock your computer,
requiring a password to unlock. Also consider logging on under only a general user account with no specific
administrative privileges, and log on with administrative rights only when you need to perform
administrative actions.

Restricting access to the server room

As discussed earlier, you should place the server room in an area that has minimal traffic. This minimizes
the opportunity to access the server room and makes unwelcome visitors more obvious. Also, be very
discriminating when determining who should see the server room. It is not a good practice to provide
tours of the server room, even to employees of the organization. Next, make sure that the door you place
on the server room is a solid, heavy, secure door made of solid wood, steel, or steel clad with a heavy duty
lock. The type of lock you choose depends upon the level of security you want or can afford. A simple
keyed lock might be sufficient; however, most organizations probably want more controlled access, such
as the following:
• Keypad: Enter a number into an electronic keypad to unlock the door. This number might be a
shared number that all IT staff knows, or it might be unique for each user. You should reset the
number on a regular basis to ensure the secrecy of the number. A keypad provides minimal
security because a passerby might be able to see a number as the user enters it.
• Card scanners: Issue authorized persons a card that is read by a scanner at the entrance to the
server room. The cards are available in a number of formats.
Some have embedded magnetic data that uniquely identifies the user. Others are smart cards that
have digital certificates with metal contacts read by the scanner.
These are nearly impossible to duplicate. Administrators can set conditions on the cards so that
they grant access to the room only at certain times—to prevent employees from sneaking into the
server room and performing malicious deeds when no one else is around, for example. When the
scanner reads the card, an electronic lock unlocks the door for a few seconds. One of the only
drawbacks to this system is that people tend to lose, misplace, borrow, or steal the access cards.
Provide a written policy to your staff that specifies penalties for missing access cards. Also make
sure that employees always wear the access card. Most cards include a photograph of the person,
helping to ensure that only the person to whom the card was issued uses it.
• Bio-recognition: This is an emerging technology that verifies a person’s identity based upon
physical identifiers such as fingerprints, retinal scans, voice imprints, or some combination thereof.

For maximum security, require some combination of access methods. For example, employees might
insert a smart card into a reader and also type a password. This method would ensure that the card was
not stolen.

48
Monitoring access to the server room
Despite your best security efforts, unauthorized persons might still find a way to access the server room,
or authorized users might damage the server room. If such unfortunate occurrences take place, utilize a
method of record keeping that allows you to know which persons were present at the time of the deed.
Some of the following methods monitor access to your server room:
• Sign-in: Clearly the least secure method, a sign-in sheet depends upon the honor system, and
persons who are a security risk are unlikely to be “honorable.” A sign-in system should implement
one or both of the next two methods that follow.
• Security guard: Because of the high cost associated with staffing a facility with a 24/7 security
guard, this option might not be practical for all environments.
However, a security guard adds an observer to unauthorized security breaches and is useful for
quick apprehension and prevention as well as adding a visible level of deterrence. A security guard
might also check contents of all bags going into the server room.
• Video surveillance: Video surveillance captures activity in the server room or at the server room
door 24/7. Some facilities keep video tapes indefinitely, but most rotate tapes on at least a seven-
day schedule. Be sure to replace tapes periodically, because older, worn tapes do not provide a
clear image. Video surveillance is only as good as the area it covers. If you cover the server room,
be sure to also cover wiring closets, areas where you store backup tapes, and so forth.
• Logs: Most controlled access methods such as scanners or electronic keypads keep logs which
show that who entered the server room and when. In addition, you can configure the NOS to track
certain resources so that if someone attempts to access a resource (whether successfully or
unsuccessfully), a log records the name of the logged-on user and time and date of access.

In securing the server room, do not forget to also secure other sensitive, physically accessible areas. You
should, for example, secure patch panels and wiring closets, which provide a point of convergence for
network cabling, making it easier to manage.

In seconds, someone could access an exposed patch panel and start ripping out cable, causing
considerable damage to network operations. Also, place backup media in secured data storage areas to
prevent stolen tapes. In the IT context, your organization is not the server room; it is the data. Stolen tapes
are a severe security risk. Secure the Telco room, which is the access point for telephone communications
and often shares its connections with a WAN or Internet connection.

Further secure the contents of the server room by placing locks on server equipment.

For example, most server cabinets require a lock, and you can optionally lock individual components of
the server rack. One reason for rack-mounted equipment is easy portability from one rack to another.
However, you do not want someone easily porting the equipment into a backpack or briefcase. Do not
leave spare parts lying around— secure them in a locked cabinet as well, and record parts with serial
numbers.

Review Questions

1. What is a network?
2. What is a LAN?
3. What is the difference between peer-to-peer and client-server network models?
4. Which is more stable, a PC or a network server?

49
5. What is the difference between authentication and authorization?
6. What are permissions?
7. What is a trust?
8. What are the three primary factors that contribute to the effective speed of the processor (CPU)?
9. Hard disk and processor cache provide what benefits?
10. How does SMP benefit the performance of a server?
11. The most likely bottleneck in most computers is?
12. Why is centralized management important regarding servers?
13. What are the three main server classes?
14. Which type of server might be cost prohibitive to a small business?
15. Why are hot-pluggable PCI slots desirable?
16. What is MTBF?
17. What is bandwidth?
18. What does load balancing do?
19. The administrator provides what to the users?
20. What is a Web farm?
21. What is a NIC?
22. Using your Web browser, access download.novell.com. Look up and read about the Novell Client
for Windows NT/2000/XP download.
23. Describe how this download helps Microsoft Windows NT/2000/XP and Novell NetWare to
interoperate.
24. Discuss some other solutions you find on the Novell site and how they work with each other. What
is e-Directory?
25. NTT India is an Internet and intranet solutions company. NTT hosts Web sites for their clients on
their servers. Of course, the servers must have significant uptime and failover protection. Access
www.nttindia.com, click the Servers link, and answer the following questions: a) What operating
system does NTT use?
b) What components on their servers are hot-swappable?
c) Describe the force-filtered cooling of NTT servers.
d) What type of redundant power system does NTT use?
e) What temperature does NTT maintain in their server room?
f) How does NTT control access to the server room?
26. Data Clean Corporation provides cleaning services for controlled environments such as server
rooms. Using your Web browser, access www.dataclean.com.
Click the services link, read the Web page, and then answer the following questions:
a) Why is it important to have the floor plenum cleaned?
b) What is important about cleaning the floor surfaces?
27. What is a plenum? Why is it useful?
28. What three power sources provide power to the server room?
29. What is clean power? Why is it important?
30. What is electromagnetic interference (EMI)?
TOPIC 2
_____________________________________________________________

2. MOTHER BOARD ARCHITECTURE, PROCESSORS,

50
 MEMORY AND BIOS
_____________________________________________________________

LEARNING OUTCOMES
After studying this topic you should be able to:
• Explain various motherboard buses and describe how clock frequency affects performance.
• Identify common server processors.
• Identify various types of memory.
• Configure the BIOS and identify common server configuration items.

2.1 MOTHERBOARD ARCHITECTURE

The motherboard provides the system “bus,” the transportation medium for data to and from processors,
memory, peripherals, and input/output (I/O) devices. Knowing about these server components helps you
to make educated decisions in the server equipment you procure. The motherboard—sometimes referred
to as the system board or backplane—is the “mother of all circuit boards.” The processor(s), memory,
buses, adapter cards, I/O ports, mass storage—just about every component in the system—are connected
directly or indirectly to the motherboard. The various hardware components attached to the motherboard
require basic management using a BIOS (basic input/output system), which identifies and confirms correct
hardware installation and configuration. You might have had at least some experience with personal
computers and might already understand many of the basic hardware components. However, when
addressing hardware relating to servers, you must be careful to change perspective. For budgetary
purposes, you might be more interested in building a server on your own instead of purchasing a
preassembled server. Experts, however, strongly recommend that you purchase servers as much as
possible as pre-tested turnkey solutions (configured hardware and operating system ready to run right
out of the box) from major server vendors. Technicians acquainted with PC workstations might be
tempted to use their technical expertise to assemble their own server systems. However, server
availability is a high priority, and even the best technician cannot responsibly guarantee the same system
uptime as major PC server vendors such as Dell, Compaq, and Hewlett-Packard (HP), which offer server
systems with between 99.9% and 99.999% uptime. (Server technicians colloquially refer to this as a certain
number of nines. For example, 99.999% uptime is referred to as “five nines” and equals about five minutes
of downtime per year.) Some vendors even promise financial compensation if your server fails.

The bus
The bus is to the server what a highway is to a transportation system. The bus provides the data path to
and from server components such as the processor and memory on the motherboard, the foundation of
the computer. The motherboard attaches to the chassis and includes slots, sockets, and other connections
for server components. A foundational architectural factor of the motherboard and its components is the
bus width, in bits. Motherboard bus width corresponds to individual data wires that transmit data. The
more wires a component such as the motherboard has, the more data it can transmit in a given period of
time. Current motherboard data bus architecture is either 32 bits wide or 64 bits wide, which you could
equate with a 32- or 64-lane data “highway.” Of course, the 64-bit data highway will be able to deliver
twice the data in the same amount of time as a 32-bit data highway. As this section will explain, a
motherboard includes a front side bus, sometimes a back side bus, and three primary I/O expansion buses:
Industry Standard Architecture (ISA), Extended ISA (EISA), and Peripheral Components Interconnect (PCI).
The speed of each bus is dependent upon the motherboard clock frequency.

51
Clock frequency
Each bus and every device that connects to the motherboard bus depends on the clock frequency of the
motherboard. Clock frequency (sometimes called the clock speed, cycle, or clock cycle) is the number of
times in one second that an electrically charged quartz crystal located on the motherboard vibrates
(oscillates). Clock frequency is measured in megahertz, a hertz equaling one cycle per second and “mega”
meaning million. If a motherboard has a clock frequency of 66 MHz, then it cycles 66 million times per
second. (Most new motherboards have a bus clock speed of 100 or 133 MHz.) The importance of the clock
speed is that the processor requires at least one cycle (and usually more cycles) for each instruction that
it executes. Therefore, the more times the motherboard clock cycles, the more instructions the processor
can perform per second. An 800 MHz Pentium processor has about 800 million opportunities per second
to perform an action, subtracting wait states in which the processor uses empty clock cycles to wait for
another instruction or hardware function to complete. Other system components, such as buses attached
to expansion slots, also depend upon the clock cycle to determine the speed with which they operate.
This unit makes no attempt to explain every type of bus, peripheral, and port, because many simply do
not apply to current technology or to servers. (Examples include VESA and MCA buses and game ports.)
However, several buses do apply to PC servers, in particular PCI buses, which are explained in this unit.

Chipset function
In early personal computers, the entire system and its components operated on one bus and ran at 4.77
MHz. This fact seems unimportant until you consider the dramatic changes in bus speeds in the computing
industry. Various hardware components require differing bus speeds in order to perform well. Faster
components can run without waiting for slower devices to complete their tasks, because the devices
operate in an independent bus context. If all components continued to run on a single bus as in early PCs,
the resulting bottlenecks would significantly defeat computing efficiency.

Using different buses in the system requires a way to divide the motherboard into separate parts. The
chipset (see Exhibit 2-1) is a group of motherboard chips that operate at the same speed as the
motherboard clock and provide the boundary that divides one bus from another and controls the flow of
bus data. Choosing a motherboard is mostly choosing a chipset.

Exhibit
2-1: The chipset is identified directly on the chip
Intel’s Web site at http://support.intel.com/support/motherboards/server displays a current listing of
server motherboards/chipsets. Intel is the primary manufacturer of PC based server boards, and
regardless of the name on the server case, most PC server vendors use Intel chipsets. Other vendors also
make chipsets that are compatible with Intel processors, but again, they tend to be low-cost alternatives
and often focus on the desktop computer market. For our purposes throughout the remainder of this
course, only systems supporting two or more processors shall be considered servers unless stated
otherwise. Also, this course does not address other chipsets such as those designed for AMD processors

52
because AMD tends to focus on desktop computers with the exception of the AMD Sledgehammer (see
the section on processors in this unit), which is developing a following in the Linux community.
Nevertheless, AMD provides excellent low-cost, high-speed processors and is making significant headway
into the desktop PC market. For more about AMD products, visit www.amd.com.

Hierarchical bus
PC-based bus systems use what is known as a hierarchical bus, because several buses actually comprise
the (collective) “bus,” each running at different speeds and with the slower buses hierarchically structured
beneath the faster buses. Dividing the bus into the front side bus, PCI bus, and ISA buses allows slower
components to operate without negatively impacting the faster components. Intel architecture utilizes a
North Bridge, South Bridge, and Super I/O chipset (see Exhibit 2-2) to divide the PCI bus from fastest to
slowest and facilitate communication between buses in the order listed:
• Front side bus—A 64-bit data pathway that the processor uses to communicate with main memory
and the graphics card through the North Bridge chipset. The North Bridge chipset divides the
processor bus from the PCI bus, and manages data traffic between the South Bridge and between
components on the front side bus and PCI bus. This core bus runs at motherboard clock speed. The
front side bus is also known by several other names, including processor bus, memory bus, and
system bus.
• PCI bus—A 32-bit data pathway for high-speed I/O for expansion adapter cards, USB, and IDE ports.
The CMOS (defined later in this unit) and system clock also connect to the PCI bus. The PCI bus
connects to both the North Bridge and the South Bridge. The South Bridge separates the PCI bus
from the ISA bus.
• ISA bus—A 16-bit data pathway for slower expansion adapter cards and the floppy disk, mouse,
keyboard, serial and parallel ports, and the BIOS via a Super I/O chip, which mitigates the need for
a separate expansion card for each of the aforementioned items. The South Bridge connects to the
ISA bus, which is the end of the hierarchical bus chain.

53
 Exhibit 2-2: Typical North Bridge/South Bridge motherboard architecture

It might be helpful for you to know the throughput capabilities of the following types of ports and buses,
because many times you will need to transfer data through these ports, even as a temporary solution. For
example, you might need to transfer diagnostic data from one server to another or to a laptop through a
special serial cable known as a null modem cable. This type of cable uses special crossed wires to simulate
a modem presence, allowing data to travel between the two hosts. In addition, many devices such as UPS
systems are now connecting to the server via a USB port instead of a serial port, and new external hard
disk storage devices can connect through a FireWire (IEEE 1394) port. The following table shows the
various interfaces and their maximum throughput from the slowest to the fastest.

Accelerated hub architecture

Intel has implemented new hub architecture to replace the tried-and-true North Bridge/South Bridge
architecture, which connects the various buses described above through the PCI bus. The accelerated hub
architecture connects buses to the system bus independently through a dedicated interface to the PCI
bus, yielding throughput of up to 266 MBps—twice as much throughput as 33 MHz PCI. The independent
buses do not share the PCI bus, thus increasing PCI bus bandwidth available to PCIconnected devices. Also,
the hub architecture improves traffic throughput between slower I/O buses and the system bus.

In an accelerated hub architecture (see Exhibit 2-3), the North Bridge is called the Graphics Memory
Controller Hub (GMCH), and the South Bridge is called the I/O Controller Hub (ICH). This architecture allows
devices directly connected to the ICH (such as high-speed ATA-66 and ATA-100 disk controllers and USB
2.0 interfaces) much greater throughput.

54
North Bridge/South Bridge architecture is common on Intel 44X series chipsets, such as the 440LX, and
the accelerated hub architecture is the current architecture in Intel 8XX series chipsets, such as the 840NX.

Exhibit 2-3: The accelerated hub architecture improves I/O traffic between slower I/O buses and the faster system bus

Bus interfaces
The front side bus, ISA bus, and PCI bus are only useful if there is a way to connect devices to each
respective bus. The front side bus uses a slot or socket on the motherboard to connect the processors and
memory. The ISA and PCI buses use expansion slots. Although PCI is quickly becoming the only slot
available on new motherboards, you might still see the older ISA/EISA slots on some motherboards, so we
briefly discuss them as well.

ISA
Industry Standard Architecture (ISA) was originally the AT (advanced technology) bus, developed by IBM
(not to be confused with the AT motherboard). Devices connect to an ISA bus through an ISA expansion
slot, which is 16 bits wide and accommodates both 16-bit devices and older 8-bit devices. The ISA bus
operates at only 8.33 MHz and is capable of transfer speeds up to 8 MBps. This performance seems slow
compared to the clock speed of the motherboard. However, most ISA devices (a modem, for example) are
even slower, so the slot performance does not hinder performance of the ISA device. You might see some
older servers with ISA expansion slots, but newer server motherboards (and workstations or home
computers) do not usually include them. Motherboards still include the ISA bus (not the expansion slots)
to accept a Super I/O chip as the means to connect slower devices such as the serial and parallel port,
floppy controller, keyboard, and mouse.

EISA
The 32-bit Extended ISA (EISA) bus provides backward compatibility with older ISA devices and a maximum
bandwidth of about 33 MBps. Even though EISA performance is better than ISA, most motherboards do
not include EISA buses anymore because most manufacturers engineer devices that are compatible with
the better-performing PCI bus interface.

55
PCI
The purpose of the Peripheral Components Interface (PCI) bus is to interface high speed devices with the
system bus so that slower devices do not create a bottleneck. For example, an older computer with a 16-
bit ISA, 8 MHz video card would create a significant bottleneck for the rest of the system (mostly the
processor) because it can only transfer a maximum of 8 MBps, and would use clock cycles that could
otherwise be used by other devices. The processor would have to wait for the ISA video card to complete
its task before being able to use the system bus. Using the bridge or hub architecture, faster PCI devices
can use the PCI bus, reducing or removing the bottleneck. Continuing the video card example, a PCI video
card uses a 32-bit 33 MHz card for much faster performance and better throughput at 132 MBps. Other
devices such as network cards and hard disk controllers can also take advantage of this improved
performance.

As for usability, one of the most significant benefits of PCI is that a Plug and Play operating system such as
Windows 95 or later can automatically detect and assign system resources to new devices. Other buses
such as ISA required BIOS configuration and/or manual configuration of jumpers and switches on devices.
PCI 2.2 is the current specification of PCI, and PCI-X is soon to supersede it.

PCI-X
PCI-X (PCI-eXtended) is actually Addendum 1.0 to the PCI 2.2 specification. The basic advantage to PCI-X is
simple: much higher bandwidth and correspondingly higher performance. PCI-X utilizes 64 bits and up to
133 MHz, yielding a maximum bandwidth of 1064 MBps. Devices that are designed according to the PCI-
X standard will be able to utilize the full maximum available bandwidth, provided no other processes or
devices contend for the same bandwidth. Also, relaxed ordering arranges real-time audio and video
instructions in an efficient order instead of the first in/first out (FIFO) method of previous PCI versions.

Other efficiency enhancements to the PCI-X bus help to free up bandwidth and reduce wait states, with
the net result of a nearly tenfold performance increase over 32 bit, 33 MHz PCI. Motherboard designers
divide the PCI-X bandwidth in one of several slot combinations for each PCI-X bus segment: one 133 MHz
slot, two 100 MHz slots, or four 66 MHz slots. The following table shows PCI and PCI-X performance
statistics.

Bus mastering
Most devices utilize the processor to control the flow of information through the bus. As a result, a
processor laden with the task of controlling requests from various devices is not as available to process
more important productivity functions, slowing down overall performance. In PCI architecture, hardware
designers can use bus mastering to bypass the processor and directly access memory, resulting in an
overall increase in processor performance. Bus mastering is actually a form of direct memory access
(DMA) known as first-party DMA. “First party” refers to the device directly controlling memory access,
and compares to a third-party DMA transfer using a motherboard DMA controller. Also, bus mastering
devices can communicate among them over the bus without CPU intervention. Video adapters and disk
controllers commonly utilize bus mastering.

PCI Interrupts
Devices issue requests for system resources using an ISA-based interrupt request. An interrupt request
(IRQ) is an electrical signal that obtains the CPU’s attention in order to handle an event immediately,
although the processor might queue the request behind other requests. Most devices utilize one of
several IRQs on the motherboard. However, there are a limited number of available interrupts, and the
number of devices is often greater than the available IRQs. The BIOS utilizes the PCI bus to assign special
56
PCI interrupts to PCI devices using the designation INTA#, INTB#, INTC#, and INTD# (sometimes known
simply as #1- #4). Single-function PCI cards always receive INTA# according to PCI specifications. Chips or
cards with multiple functions can receive assignments for INTB# through INTD# as needed. The PCI
interrupts map to one of four corresponding ISA IRQs, usually IRQ 9-12. For example, if you have three
single-function PCI cards, they all receive INTA#; however, each device still requires a unique ISA IRQ
mapping. Functionally, the result is not that much different than if each device was a standard ISA device
in the first place, because each device still receives unique, non shareable IRQs. The benefit of the PCI
interrupt appears when no more ISA IRQ addresses are available. With no more available IRQs, the PCI
interrupt utilizes another PCI function known as PCI steering, in which the PCI interrupt assigns two or
more PCI devices the same ISA IRQ.

PCI hot swap

PCI hot swap, otherwise known as PCI hot plug, means that you can add, remove, or replace PCI devices
without first powering down the server. Note that even though it sounds like you can just take off the
cover and rip out a card, many devices and PCI slots require you to follow specific steps. For example,
most servers require you to turn off the power to the slot using management software (as with many Dell
and HP systems) or using a switch or button (as with Compaq systems) before removing the device (see
Exhibit 2-4). The power switch is often a button located near the actual slot. After you turn off the power
to the slot, then remove and replace a card, you can turn the power back on. If the NOS is Plug and Play
compatible (such as Windows 2000), it should be able to detect the new device and load (or request) the
appropriate drivers. Hot swapping is truly a lifesaver in servers that require 24/7 operation.

Exhibit 2-4: To turn off power to a PCI slot, use a button (as shown) or management software

Peer PCI bus

The peer PCI bus is usually a server-specific function that both increases available PCI bandwidth and
expands the number of PCI expansion cards from the usual limit of four with a minimal impact on overall
system bus bandwidth. This architecture usually involves dual peer PCI buses and two North Bridges,
which connect to a primary PCI bus and a secondary PCI bus. PCI expansion slots connect to each
respective PCI bus and are either integrated into the motherboard or installed as add-on daughter cards.
Many motherboards use this expanded functionality not only to increase the number of expansion slots,

57
but also to offer flexible PCI bus width and speed. For example, Bus #1 could offer four standard 32-bit 33
MHz PCI slots, while Bus #2 offers two additional
64-bit 66 MHz PCI slots and devices in both buses can simultaneously access their respective buses. Peer
PCI slots allow the administrator to load balance the system. For example, if you have two high-speed
network cards for which you expect a great deal of traffic, you could place each one on a separate PCI bus
to balance the load.

That way, they can each handle I/O without waiting for the other to complete a task on the PCI bus. You
can extend the same load-balancing benefits to other devices such as high-throughput SCSI controllers.
High-end servers such as the HP Net server running Windows 2000 Datacenter Server offer up to 32
processors and 96 PCI slots!

Compare peer PCI slots with the bridged PCI bus, in which an additional PCI-to-PCI bridge is inserted below
the North Bridge. This only increases the number of available slots, but does not offer better PCI load
balancing because all expansion slots actually use a common data path to the system bus. In the peer PCI
architecture, separate buses independently communicate with the system bus for more efficient load
balancing.

I2O
Intelligent Input/Output (I20) is an initiative to improve I/O performance via an I2O processor and driver
model. The I2O driver communicates with the I2O processor, which is located on the device itself, as a
separate add-in card, or integrated into the motherboard. Even on the PCI bus, which is designed to relieve
traffic from the system bus, frequent PCI interrupts to the processor slow overall performance. With I2O,
devices intelligently perform much of the processing function on their own. Also, I2O devices can
communicate among themselves when necessary instead of using the processor to manage their
communication. The I2O driver utilizes a “split driver” model in which the Operating System Module
(OSM) handles I/O interaction between the device and the operating system, and the Hardware Device
Module (HDM) manages interaction between hardware controllers and I2O-compatible devices. The I2O
specification goes a long way toward developing a common standard that hardware and software vendors
can use to simplify and reduce the costly, time-consuming process of driver development. Most I2O-
enabled devices are network cards or storage devices because they typically require the highest I/O levels
in the system. The I2O specification can work with OS/2, but don’t expect to see great strides in I2O
technology on the OS/2 platform. Windows NT 4.0 initially offered no participation with I2O; however,
Windows 2000 and NetWare 5.x fully support the I2O specification.

The latest I2O specification (2.0) includes several new features, the most significant of which are:
64-bit addressing accommodates increased memory capabilities for newer CPUs.
• Hot-plug capability lets you change the adapter without shutting down the entire system.
• Direct memory access (DMA) allows direct access to memory instead of first utilizing the processor.
• High-speed I2O RAID disk controllers in particular benefit from DMA.

Accelerated Graphics Port (AGP)

The Accelerated Graphics Port (AGP) is designed to relieve the system bus and CPU of traffic and
processing. Producing graphics is a very complex function, requiring memory usage and significant
processing power. In the past, graphics functions used portions of main memory and depended on the
main CPU to process much of the graphics load. Later, as video adapters matured, they performed much
of their own graphics processing by adding memory chips and onboard processors specially designed for
graphics functions. Nevertheless, the graphics card would frequently request attention from the system
58
bus and the processor. The AGP specification introduced in 1996 utilizes a single AGP slot on standard
motherboards, and further relieves the processor and system bus of video burdens. The slot is brown in
color and fits only AGP cards—so you can’t accidentally insert any other type of card into the slot.

The initial specification offered both a 1X and 2X mode, representing a performance multiplier of 2,
doubling the effective clock speed of a 1X card. Later, a 4X mode appeared; however, most cards are still
produced at the 2X speed. There is also an AGP Pro spec, which uses a longer slot and more pins for higher
voltage. In November 2000 the AGP 8X was introduced, but at this writing there are no cards available for
it. The following table lists AGP performance statistics.

AGP has an immediate and obvious benefit to overall system performance; however, AGP provides the
greatest benefit to graphics-intensive computing, such as PC gaming, computer aided drafting (CAD),
graphic design, and other high-end graphics applications. Some servers come with AGP, particularly dual-
processor machines that could just as easily serve as high-end graphics workstations. High-end servers
normally do not include AGP because it is not necessary and adds a potential point of failure.
Manufacturers try to ensure highest availability for servers by not including complex graphics features.
You are not likely to be playing PC games on the server, so there is really no need for AGP graphics. High-
end servers usually include a motherboard-integrated video adapter at 1024 X 768 screen resolution and
only 256 colors. By graphics standards, this is video from the late 1980s.

However, its simplicity avoids potential graphics problems on the server. Also, it does not matter from the
administrator’s perspective that the graphics are unimpressive, because most day-to-day server
administration is actually done remotely on a desktop PC workstation.

PCI-Express
PCI-Express, formerly called 3GIO (third generation I/O), is designed to replace PCI, PCI-X, and AGP.
PCIExpress is not just an upgraded PCI; it’s based on a different model. Rather than using a 32- or 64- bit
shared bus, PCI-Express uses a point-to-point serial connection with each device or slot. Packets of data
are sent over transmit/receive pairs, much like an Ethernet connection. The theoretical maximum speed
of one PCI Express Serial pair, or “lane,” is 2.5 GB/s each way, or 5GB/s total. The actual performance is
slower, but a device, such as a network card, using a single lane is still at least twice as fast as a regular
PCI device. Furthermore, devices in need of more bandwidth can have several of the individually-clocked
lanes. For instance, the PCI Express replacement for AGP is a 16-lane, or x16, slot. Because the link is serial
and not parallel, PCI-Express requires smaller connectors. An x1 device has 36 pins, compared to the 120
required in a regular PCI slot. A system board can have a mixture of PCI-Express slots, including x1, x4, x8,
and x16 slots. A device can fit into a slot that is larger than it needs – the extra contacts will be ignored.
Although older PCI devices won’t fit in PCI-Express slots, compatibility with PCI addressing, and thus
drivers and software, has been maintained. Other features that will be available to operating systems and
applications include advanced power management; hot swap/hot plug ability, and data integrity and error
handling.

2.2 PROCESSORS

It is not within the scope of this course to exhaustively describe every processor known to the PC world,
starting with the 8088 pioneer of the Intel platform and finishing with the latest Pentium. You should,
however, be aware of the characteristics of common processors found in servers today. Another reason
for not covering earlier processors is that you are less likely to find older processors in servers, because
administrators tend to update server processing power more frequently than desktop workstation
59
processors. This makes sense because business applications and data on servers become larger and more
complex much faster than on desktops, and because server performance affects multiple users.
Therefore, a server upgrade yields a higher return on investment than a workstation upgrade, which
benefits only a single user. Also, this section focuses on Intel processors for good reasons. First, PC servers
by definition involve Intel or Intel based processors. Otherwise, they would be something like a RISC
processor. Also, you will find Intel processors in more PC servers. An Intel competitor, AMD, also makes
Intel-compatible processors. However, AMD largely aims its efforts at cost-effective desktop PCs and
workstations.

We will discuss AMD to some degree in relation to servers because its latest product, a 64-bit processor
is aimed at high-end workstations and low-end servers. In fact, Compaq has started to make servers using
the AMD processor.

Processor speed
Processor speed is a measure in MHz of the number of opportunities per second that the processor can
execute an action. Recall that each clock cycle represents an opportunity for the processor to do
something. The processor architecture design uses a multiplier methodology to provide the processor’s
speed. For example, a Pentium III 600 MHz processor installed on a motherboard with a system bus speed
of 100 MHz uses a multiplier of 6 (100 MHz system bus speed X multiplier factor of 6 = 600).

Cache memory
Cache memory is a small amount of memory that stores recently or frequently used program code or data,
reducing the latency involved in retrieving data from RAM or disk. Cache memory appears in a number of
places on the server, including the hard disk, CD-ROM, and processor. Processors use two types of cache
memory: L1 (level 1) and L2 (level 2).

L1 cache
L1 cache is a small amount of memory (usually 32–64 KB) that provides extremely fast access to its data
because of its proximity to the processor and because it runs at the same speed as the processor itself—
not at the speed of the motherboard. For example, a Pentium III 850 running on a 100 MHz motherboard
utilizes an L1 cache that also runs at 850 MHz, not 100 MHz. L1 cache provides an advantage to system
performance, because the processor can access data directly from the L1 cache instead of having to fetch
the data from memory, which is slower, or from the hard disk, which is painfully slower. Also, if the data
in the L1 cache is the result of a processing action such as a complex calculation, retrieval from the L1
cache conserves valuable processor utilization because a recalculation is not necessary. While the size of
the cache seems too small to be of any use, it is a great benefit because frequently used chunks of code
or data are constantly served from extremely fast L1 cache.

L2 cache
L2 cache provides the same basic benefits as L1 cache, but it is larger, ranging from 256 KB to 2 MB. In the
past, L2 cache was not stored on the processor die, but was instead stored on a separate chip inside the
processor housing. This orientation is known as discrete L2 cache. The data path used to access the L2
cache was called the back side bus, and it ran at half the processor speed. For example, a Pentium III 450
MHz processor utilizes a 512 KB L2 cache running at 225 MHz. Some Pentium III 500s run a 512 KB L2
cache at half the processor speed and some run a 256 KB L2 cache at full processor speed. Most processors
after the Pentium III 500 locate the L2 cache directly on the processor die (similar to the L1 cache) and run
it at full processor speed. This Advanced Transfer Cache (ATC) is 256 bits wide and eliminates the need for
a back side bus.
60
 Exhibit 2-5: L1 cache, L2 cache, and Advanced Transfer Cache

Server processors
In existing servers, you are likely to find processors no slower than Pentium II 233 MHz or Pentium Pro
200 MHz. Servers also utilize an additional processor model not typically found in home or standard
desktop PCs—the Pentium II Xeon and Pentium III Xeon (pronounced “zeon”; see Exhibit 3-6). Xeon
processors differ from standard Pentium II or III models in the following respects:
• Type of enclosure
• Cache size
• Cache speed
• Amount of addressable memory
• SMP (symmetric multiprocessing)

The Xeon uses a Slot 2 single-edge contact (SEC) enclosure and is larger than a Pentium III in order to
accommodate the internal board with more L2 ATC memory— up to 2 MB of error checking and correction
memory. The Xeon uses a 256-bit data path to the L2 cache—a fourfold improvement over the standard
Pentium II/III 64-bit data path. On the Xeon, both the L1 and L2 cache can run in parallel, offering
simultaneous access and further reducing latency. The Xeon increases the number of fill buffers, the
interface between the CPU and main memory, from four to eight, and increases bus queue entries, which
hold outstanding bus and memory operations, from four to eight.

The amount of memory that the processor can use is a factor of the processor bit width and motherboard
chipset. Typically, the processor can address more memory than the motherboard allows. A 32-bit
processor can address 4 GB of memory (2^32 = 4,294,967,296 bytes, or 4 GB). In the home desktop and
corporate workstation, you are unlikely to find a motherboard with physical space and chipset design to
allow for this much memory. Server processor design, however, is changing to allow substantial amounts
of addressable memory by modifying the motherboard and/or chipset. Intel Pentium II Xeon and later
processors let the processor utilize 36 bits to address memory using Intel’s Physical Address Extension
(PAE) feature, allowing up to 64 GB of addressable memory (2^36 = 68,719,476,736, or 64 GB). While even
4 GB sounds like an immense amount of memory (and it is), large, real-time server applications such as
online transaction processing (OLTP) and e-commerce require large amounts of data to reside in RAM for
fast access. A Pentium III motherboard configuration accepts either single or dual processors.

A Xeon SMP configuration can use up to four processors, though by adding another processor bus (often
called a mezzanine bus), eight processors are possible, and some manufacturers engineer buses that can

61
use up to 132 processors (though four or eight is more common). A dual-processor system is known as “2-
way,” four processors as “4- way,” eight processors as “8-way,” and so on.

Notably absent from the general discussion to follow is the Celeron processor, because it is designed for
the low-cost home PC market. Intel reduces the cost by utilizing a smaller cache and cheaper packaging,
although the core Celeron II/III is the same core as the basic Pentium II/III. Also absent are the Classic
(original) Pentium and the Pentium MMX because it is unlikely that you will find these in servers—
although MMX video technology is still present in server processors. The Pentium 4 processor offers
significant performance benefits over the Pentium III; however, it is not capable of SMP and is geared
toward the high-end workstation or demanding home user.

64-bit processors
As with motherboard buses and adapter cards, the bit width on a processor correlates to the amount of
data that it can transmit. Each bit corresponds to a wire connector through the socket or slot for data
transmission between the processor and the motherboard.

Most Pentium processors function internally at 64 bits, and then the data results are passed on to the 32-
bit external bus interface.

Most server processors use 32-bit bus interfaces, but new processors from Intel and AMD are 64-bit
processors both internally and externally. Although these processors are not in final form at this writing,
they are likely to affect the future of PC server computing very soon.

Intel Itanium
The 64-bit Intel Itanium using Intel’s IA-64 technology represents a departure from 32-bit x86 Intel
architecture, and performs optimally with 64-bit operating systems and applications (all editions of
Windows Server 2003 have 64-bit versions). Co-developed with HP, the Itanium depends upon new
compiler technology. (A compiler translates a high-level programming language into the lowest language
the computer can understand, machine language.) In addition, 32-bit applications running on the Itanium
processor utilize the Itanium’s hardware emulation to adapt the 32-bit instructions for the 64bit
architecture. Because of the translation process, 32-bit applications will usually run more slowly on the
IA-64 than on fast 32-bit Pentium III Xeon processors.

The Itanium processor runs on Intel’s upcoming 460GX chipset.

One of the reasons many large organizations will migrate to the Itanium 64-bit platform (IA-64) is not so
much the core processor speed as the 64-bit memory addressability.

With 64 bits, the processor can address up to 18 billion GB (2^64 = 18,446,744,073,709,551,616 bytes).
This seems like an absurd amount of memory, but at least it doesn’t appear as if there will ever be a
memory ceiling again, and memory hungry applications such as databases will make good use of any
available memory. In addition to a large L2 cache, the Itanium also supports a 2 or 4 MB Level 3
motherboard cache (much like L2 cache before ATC).

The IA-64 architecture uses Explicitly Parallel Instruction set Computing (EPIC), allowing the processor to
simultaneously process as many as 20 operations. New motherboard designs will take advantage of IA64
architecture to also allow handling of up to 64 bits of data at a time. This type of functionality will be an
especially powerful feature when applying 64-bit processing to encryption schemes such as RSA
62
encryption/decryption. Intel estimates that an IA-64 processor will outperform the fastest RISC-based
processors by a factor of eight or more.

Motherboard architecture, processors, memory, and BIOS 3–19

AMD Sledgehammer
The AMD Sledgehammer (built on the AMD Athlon core) is also a 64-bit processor, but there is otherwise
little similarity between the Itanium and the Sledgehammer (or “Hammer”). AMD decided to extend
Intel’s original x86 architecture in the Hammer design with AMD’s new x86-64 architecture. In fact, 32-bit
operating systems and applications can run on the Hammer without complicated hardware emulation,
resulting in minimal performance overhead. This strategy could prove to be a wise marketing move for
AMD, because few enterprises will be able to switch all operating systems and applications to 64-bit
overnight. The Hammer allows organizations to gradually merge 64-bit functionality into their existing
framework. Unfortunately, Microsoft has committed to creating 64-bit operating systems and
applications only for the Intel IA-64 platform, not the AMD x86-64 platform. The Linux community,
however, already has a 64-bit version of Linux under way, and Sun Microsystems has also announced that
it will port Solaris UNIX to the x86-64 platform. AMD is developing a new Lightning Data Transport (LDT)
system bus with throughput as high as 6.4 GBps.

2.3 MEMORY

Many types of memory have been available in PCs and servers over the past few decades, but this section
describes only the types of memory most likely to be found in servers today. Desktop PCs and servers
share many of the same memory characteristics, but servers often have additional memory features, such
as registered memory and ECC memory. Several dynamic RAM (DRAM) memory chips are installed on a
printed circuit board (PCB), which is collectively referred to as a module. DRAM is dynamic random access
memory—referred to as dynamic because the information requires continuous electrical refresh, or else
the data can become corrupt or lost.

SIMM modules
You will probably find SIMMs (single inline memory modules) only on older servers. The original SIMM
was 8 bits wide (plus one optional parity bit) and used 30 pins to connect to its slot on the motherboard.
However, you won’t find these older modules in servers. Instead, you are more likely to find SIMMs with
a data path of 32 bits (plus four optional parity bits) with 72 pins (see Exhibit 2-7). The physical SIMM
module has gold or tin contacts at the bottom. Although the contact appears both on the front and back,
it is really a single contact (hence the single in SIMM).

SIMMs is rated according to the time it takes to retrieve data from memory in nanoseconds (ns), which is
one billionth of a second. The typical SIMM is 50 ns, 60 ns, or 70 ns, and older SIMMs can be 80 ns. Even
memory running at 60 ns is currently considered slow in light of faster SDRAM DIMMs,

Exhibit 2-7: A 72-pin SIMM

63
EDO
SIMMs and early DIMMs are known for the Extended Data Out (EDO) RAM technology (sometimes called
“hyper page mode”), which relates to locations in memory known as memory addresses. A memory
address references rows and columns.

Instead of providing only the exact location requested, EDO can send the entire row address so that
subsequent references to the same row require only a column lookup, saving time. This functionality is
the same as an older technology known as Fast Page Mode (FPM) RAM, and also adds the ability to
eliminate a 10 ns delay prior to issuing the next requested memory address.

DIMM modules
DIMMs (dual inline memory modules) dramatically improve memory performance over SIMMs by
expanding the module to 64 bits (non-parity) or 72 bits (parity or ECC) using 168 pins (see Exhibit 2-8). The
contacts on both sides of the module are separate (hence the dual in DIMM). Recall that the more bits
available for the data, the more data that can be processed in a given period of time. Because a DIMM
uses 64 bits (instead of 32, like a SIMM), it yields a performance increase.

Exhibit 2-8: A 168-pin DIMM

SDRAM
Closely associated with a DIMM is SDRAM (synchronous dynamic RAM), because a DIMM is the physical
platform of SDRAM. SDRAM removes the FPM and EDO DRAM signal-controlled bottleneck that emerged
as buses faster than 66 MHz appeared. At 60–80 ns, the processor would request information faster than
memory could serve it over the bus. SDRAM operates at clock speed; if the system bus is 100 MHz, then
SDRAM matches that frequency, which functionally operates at about 10 ns. SDRAM memory is referred
to with a PCXXX, where XXX is the bus speed for which the memory is designed. For example, PC100 refers
to memory designed with a rated speed for use in 100 MHz motherboards. However, manufacturers
actually make SDRAM run at 125 MHz on a 100 MHz bus for added margin because of extremely tight
nanosecond timing. SDRAM is also available at 133 MHz at about 7.5 ns. Expect even faster SDRAM to
match increasing speeds of newer buses. To identify the speed in nanoseconds upon visual inspection of
the memory chip, look at the digits at the end of the product number. You should see 10 for 10 ns, 8 for
8 ns, and so on.

RDRAM
RDRAM (Rambus DRAM) is an invention of Rambus Technology. Rambus does not actually manufacture
memory, but it developed the technology and charges royalties against memory manufacturers. RDRAM
memory chips fit on a narrow, 16-bit-wide RIMM memory module. (RIMM is not an acronym; it’s a
Rambus-patented name). RDRAM provides extremely fast 800 MHz internal clock speed on a 400 MHz
bus, because data is transferred on both the leading and trailing edge of each clock cycle.

64
This adds up to 1.6 GB throughput (16 bits X 800 MHz / 8 = 1.6 GB). Intel has expressed the most interest
in RDRAM, making it the memory of choice in the 820 chipset for PC desktop platforms, the 850 series for
Pentium 4 platforms, and the 840 chipset for high-end workstation and server platforms. However,
because of the licensing royalty and tight production tolerances, other chipsets (such as AMDbased
chipsets) avoid RDRAM, preferring DDR SDRAM instead. The RDRAM data path must travel through each
RIMM from beginning to end, which adds a delay when data exits the modules. Compare this to DIMMs,
with parallel connections to the motherboard, which allow independent data throughput for each DIMM.
Because of the unique data circuit of RDRAM, empty RIMM sockets must be filled with a C-RIMM, a device
that has no memory but provides continuity to complete the memory data path.

The RIMM is uniquely identifiable because you cannot see the actual memory chips, as with other memory
types. Instead, an aluminum sheath known as a “heat spreader” covers the RDRAM to help diffuse high
heat levels brought on by the fast access and transfer speeds.

DDR SDRAM
Double data rate SDRAM (DDR SDRAM) is the next generation of SDRAM, and also uses a 64-bit DIMM
with future plans for a 128-bit DIMM. DDR SDRAM (or SDRAM II), like SDRAM, is synchronous with the
system clock. However, DDR SDRAM transfers data twice per clock cycle, similar to RDRAM, but at a lower
cost because DDR SDRAM is an open standard charging no royalties. If the bus is 133 MHz, DDR SDRAM
transfers data at 266 MHz. In addition, it retains the data pathway of DIMMs, offering faster data transfer
from the actual DIMM to the bus with parallel construction, as opposed to the continuity requirement of
RDRAM.

DDR-2
DDR-2 is similar to DDR, with some improved features, such as 4-bit prefetch (up from 2), enhanced
registers, and on-die termination. DDR-2 requires 240 pins, as opposed to DDR’s 184. Both use a 64-bit
interface. DDR-2 operates at 1.8 volts, about half that of DDR. The first examples of DDR-2 memory do
not actually outperform existing DDR memory, but the technology will allow for improvement in the
future.

Interleaving
Interleaving allows memory access between two or more memory banks and/or boards to occur
alternately, minimizing wait states. For interleaving among banks on the same board, you must completely
fill the first bank, and then completely fill the second bank with memory that is identical in size and speed.
For example, if you have two banks of memory with four slots each, Bank A and Bank B, and Bank A has
256 MB RAM in each slot totaling 1 GB, and then Bank B must have exactly the same memory
configuration.

65
 Exhibit 2-9: Interleaving between banks of memory—Bank A and Bank B must be filled identically

If you use larger memory configurations, such as with servers that have separate dedicated memory
boards, you can interleave not only among banks on a board, but also among the boards. This
configuration also requires you to configure RAM pairs identically. To interleave boards, each pair on one
board must exactly match the corresponding pair on the other board (see Exhibit 2-10). See more about
your vendor's specific interleaving requirements. For example, HP has several other considerations for
their Net server Lxr8000 servers (www.netserver.hp.com—search for board-to-board interleaving).

Exhibit 2-10: Interleaving between banks of memory—corresponding pairs between boards must be filled identically.

Interleaving configurations are described in an X-way format, where X is the number of interleaved banks
in use. For example, if you have two memory boards with two memory banks, you have four-way
interleaving (2 boards X 2 banks = 4-way).

Buffered and registered memory

Buffered memory is a function of FPM or EDO memory, and is an older memory technology. The purpose
of both buffered and registered memory (which is more common in current computers) is to re-drive
(amplify) the signal entering the module.

Buffered or registered modules, which have a synonymous function, also assist the chipset in handling the
larger electrical load when the system has a lot of installed memory, allowing the module to include more
memory chips, which is one reason that servers often use registered modules. (With desktop PCs, SIMMs
66
are likely to be un-buffered, because the chipset manages the buffering function.) You find registered
memory on servers or high-end workstations, but rarely on a desktop PC. Registered memory also enacts
a deliberate pause of one clock cycle in the module to ensure that all communication from the chipset
arrives properly. Registered memory is useful on heavily loaded server memory, and was designed for
SIMMs containing 32 or more chips.

Error Correcting Code (ECC)

Servers commonly use error correcting code (ECC) SDRAM. Although error correction is more expensive
and involves a slight performance penalty, it is well worth it on a server, where data integrity is critical
and other high-performing system components help make up for memory latency. ECC calculates check
bits and appends them to the data during memory writes. For memory reads, ECC decodes the appended
check bits and compares the write and read check bits. If there is a discrepancy in the check bits, then an
error has occurred, and the NOS can be notified. If only a single bit error occurs, ECC can correct the error,
but ECC cannot correct the more rare 2-, 3-, or 4-bit errors (multiple bit errors). To find out more about
specific memory modules, or to order memory for your servers, check the following Web sites:
• www.crucial.com
• www.micron.com
• www.kingston.com
• www.pny.com

2.4 CMOS, BIOS, AND POST

The computer needs to have a way of finding its bearings—that is, it requires a means to locate, identify,
and configure the various hardware components in the system.

Hardware settings apply to two basic stages: first, when you turn on the system and, second, when the
NOS loads. The NOS detects and/or applies configured resources to system hardware. This section
addresses how the CMOS, BIOS, and POST relate to configuring and detecting hardware when the system
is powered on.

CMOS
The CMOS is a complementary metal oxide semiconductor that includes a small amount of memory, the
purpose of which is to store the BIOS settings such as the boot order (floppy, CD-ROM, hard disk, and so
forth), hard disk configuration, power management settings, and more. The CMOS can store data for as
long as power is available. The power supply provides power when the system is on; and a small,
nonrechargeable, metal oxide battery (similar to a watch battery) supplies power when the server is off.

Unlike desktop PCs, servers are usually powered on continuously except for regular maintenance,
hardware upgrades, and troubleshooting when you must necessarily power off. Typical batteries can last
for more than five years on desktop PCs, and somewhat longer on servers since the battery charge
naturally dissipates as a matter of time instead of actual use. Server replacement might precede battery
replacement, so sometimes battery life is not an issue. However, you should stock a few of the most
common batteries in use on your servers just in case. You identify the specific battery in use by reading
the identification stamped into the battery surface. Batteries near the end of their life usually lose time
on the real time clock, so you should replace batteries on systems with slowing time.

67
BIOS
The Basic Input/output System (BIOS) is a series of software programs that is the lowest-level interface
between the hardware and the operating system. The BIOS programming is stored on a flash BIOS memory
chip, also known as EEPROM (electrically erasable programmable read-only memory). The administrator
can configure the BIOS programming to suit his or her needs and preferences, and the configuration is
stored in the CMOS, which is powered by a small battery that retains the settings even when the power
fails or is turned off. As its name implies, the BIOS is a series of input and output configuration settings for
peripherals, adapters, and onboard components. Phoenix Software, Award Software, and American
Megatrends Inc. (AMI) create most base BIOS programming, though individual server manufacturers often
add modifications to provide functionality with their specific hardware. Phoenix Technologies acquired
Award Software in late 1998, so a newer system will usually use either a Phoenix or an AMI BIOS. The BIOS
controls all of the hardware on the system board and acts as a bridge for various NOS hardware drivers.

Accessing CMOS settings

Most of the time, when you turn on the power, the display tells you a specific key or keyboard combination
to press in order to access the BIOS settings in CMOS.

Typically, this is F1, F2, Esc, Del, or some combination of Ctrl+Alt (such as Ctrl+Alt+Shift, Enter, Esc, or S).
Compaq computers usually use F10. A simple instruction such as “Press F2 to enter settings” often appears
on the screen. Several manufacturers display a manufacturer-promoting splash screen that would prevent
you from seeing such instruction, although pressing the Esc key often removes it. (In the unusual instance
that the display does not indicate a method for accessing the BIOS, you can trick the system into allowing
you to access the BIOS by pressing and holding virtually any key immediately after you power on the
computer. The BIOS will often interpret your action as a keyboard problem and provide an opportunity
for you to access the BIOS.)

When the system powers on, a procedure known as the POST (power-on self-test) verifies functionality of
motherboard hardware. If the settings do not match, one or more beeps occur. Check system
documentation to interpret the meaning of beep codes, which usually also accompany an on-screen error
notification code. During the POST, if a device has its own BIOS such as a video card or SCSI card, the POST
allows the device to perform its own diagnostics and then resumes when the diagnostics are finished. The
POST checks the following:
• Video card and monitor
• CPU stepping (specific incremental version of the CPU)
• CPU model and speed
• BIOS version
• RAM
• Keyboard (which it enables)
• Various ports such as USB, serial, and parallel
• Floppy and hard disk drives
• Disk controllers using separate BIOS
• CD-ROM or DVD-ROM
• Sound cards
• Operating system (which it finds and loads)

68
Protecting the CMOS
Protecting the CMOS for both the server and workstation is an important security precaution. For
example, anyone with physical access to the server could access the BIOS settings in CMOS to ensure that
the system can boot from a floppy disk. Then, after booting from an MS-DOS or Windows 98 boot floppy,
he or she could gain access to local hard disks and steal, alter, destroy, or otherwise damage data or the
operating system. Protecting the CMOS first involves physical security, and then applying a password to
the CMOS. The BIOS menu system is usually easily navigable, and you should be able to locate where to
designate a password for the BIOS settings in CMOS. The CMOS usually includes two levels of password
protection: a password to access and change the BIOS configuration, intended to prevent the curious from
viewing the CMOS settings, and a password to boot the system. After setting the password(s), be sure to
record and store them in a secure location.

In the event that you cannot find a server’s CMOS password, you have no choice but to reset the CMOS,
which clears password settings in addition to any configuration settings. Reset the CMOS in one of two
ways. First, use jumper pins to short the battery circuit to the CMOS. To find the exact jumper pins, refer
to the motherboard manufacturer’s manual. In absence of a manual, you can also search for labeling on
the circuit board. Usually the label is something like CPW (clear password), RPW (reset password), or the
unmistakable “Short here to clear CMOS.” Without battery power, BIOS settings drain from the CMOS and
the password resets to the original null (none) password setting. Second, you can simply remove the
battery for a few seconds to clear the BIOS configurations, and then replace it. Either way, when you
reboot, there will be no password required to boot or access BIOS settings.

If you clear the BIOS settings, the operating system, hardware devices, or system preferences such as
power conservation settings might not function as expected.

Although a BIOS maker provides the basic BIOS, the server manufacturer has likely shipped the computer
with default settings. Without these default settings, items such as the hard disk might not be accessible,
making a boot impossible. After resetting the CMOS (or after you replace a dead battery), navigate the
menu system to locate a setting that restores default settings to get back to a workable starting point.
(Fortunately, the BIOS interact with several hardware components to automatically configure settings, as
in the hard disk configuration and memory detection.)

Common BIOS settings

It would be impractical to list all common BIOS settings from all manufacturers. A general list of
configuration categories usually revolves around the system time/date, hard and floppy disks, disk
controller (IDE), keyboard, processor(s), memory, ISA/EISA/PCI expansion buses, system resources (IRQ,
DMA, etc.), serial/parallel ports, and boot configuration. However, you should know some common
features that you are more likely to find on a server than a workstation. After you access the BIOS settings
in CMOS, use the menu system to navigate to various settings. The BIOS menu system varies from one
manufacturer to the other, and BIOS from a particular BIOS maker might be different from one machine
to the next because server manufacturers often modify the menu system for uses specific to their
computers.

Most servers also include an additional management utility that provides similar management to that seen
in the BIOS, except using a manufacturer-specific interface and settings. The settings are usually saved in
BIOS, but settings can go beyond basic BIOS settings. For example, Intel’s System Setup Utility also allows
you to save field replaceable unit (FRU) information. (An FRU is a system with replaceable

69
CPU, CMOS, CMOS battery, RAM, and RAM cache.) Typically, the management utility is a DOS-based utility
run from floppy disk(s) or a CD-ROM. In some server types, however, these devices are not replaceable
without sending the server to the manufacturer.

Review Questions

1. How does the bus bit width affect data throughput?

2. If a server offers uptime of five nines, how many minutes per year might it be down?
3. How does clock frequency affect server performance?
4. What is the function of a chipset?
5. What is a hierarchical bus?
6. How does PCI-Express differ from previous PCI versions?
7. What is the purpose of L1 and L2 cache?
8. Which characteristics differentiate the Intel Pentium III Xeon from the Pentium III?
9. What is an advantage of the Intel Itanium?
10. Using your Internet browser, access http://developer.intel.com and click the link for server
components and products, then click the link for server boards. Choose a model to research. a)
How many processors does this board support?
b) What kinds of processors are supported?
c) What kind of memory is used?
d) What is the maximum amount of memory supported?
e) What are some ideal applications for this board?

70
 TOPIC 3
_____________________________________________________________

3. SERVER POWER AND RACK INSTALLATION

_____________________________________________________________

LEARNING OUTCOMES
After studying this topic you should be able to:
• Identify features of server power supply and correctly implement an uninterruptible power supply
(UPS).
• Plan optimum placement of equipment in a server rack, configure a keyboard, video, mouse (KVM)
console, and list tips for installing equipment in racks.

3.1 SERVER POWER

Power is an obvious requirement for a server, and this unit shows you how server-level power supplies
differ from standard desktop workstation power supplies. While a server power supply typically has a
respectable mean time between failures (MTBF) of around two million hours, power to the office building
does not. Many factors can cause power to fail to the server room, and you should prepare server room
equipment for such an event with one or more devices that can temporarily provide power. The
byproduct of powering the server is the generation of heat. Fans are the primary method of protecting
server components from overheating. It’s almost impossible to install too many fans in the server, rack,
or cabinet; and this chapter shows you various types of fans and how to strategically place them. This is
especially important in the rack, where multiple server devices compound heating issues. Installing server
equipment in the rack is a strategic process that requires planning in terms of power, heat, and weight
distribution. In fact, the server rack often has so many installed servers that it is impractical to attach a
separate keyboard, monitor, and mouse to each server. In that case, you will need to use a single
keyboard, monitor, and mouse to switch between numerous servers.

Supplying power to the server

Obviously, without power you don’t have a working server. Selecting a server with the type of power
supply you need and ensuring its continued operation through the use of an uninterruptible power supply
(UPS) is the first step in ensuring solid uptime percentages.

Power supply
On the most basic level, an entry-level server probably has one power supply of at least 330 watts (W),
whereas most workstations are probably as low as 145 W. A high-end, 8- way server probably provides
between two and four power supplies rated between 375 W and 750 W each. The power supply, also
known as the power supply unit (PSU), attaches to the server chassis. Low-end servers and high-end
workstations might not have hot-swappable and/or redundant power supplies. This means that in order
to replace the power supply, you must remove the server case cover and disconnect power supply
connectors to the motherboard, hard disks, floppy disks, and CD-ROMs before removing the screws
attaching the PSU to the chassis and lifting out the power supply.

This type of power supply installation is no different from the procedure on a typical desktop workstation
because the chassis includes space for only a single, non- hot swappable PSU. Wellconfigured servers, on

71
the other hand, offer at least two hot swappable power supplies. You can replace one of the power
supplies without turning off the server.

Typically the server can operate acceptably with only a single PSU. The second power supply provides load
balancing to reduce demands placed on a single PSU, as well as failover in case one of the power supplies
fails. Often, a server includes three power supplies—two to provide continuous power and a third on
standby in case one of the first two fails. Server management software and warning lights on the server
should alert you to failed or unstable power supplies, usually indicated by a fan that does not spin at
appropriate levels (see Exhibit 3-1).

Exhibit 3-1: Most PSUs include a warning light that notifies you of problems

If one PSU fails, a standby comes online automatically and without interruption in service. (Some server
configurations such as the Intel AD450NX server platform include support for an optional fourth PSU.)
Because most networks require server availability 24/7, PSU failover is critical to a true server-level
configuration. Also, the hot-swappable power supply does not require separate connectors for hard disks,
CDROMs, the motherboard, and so forth. A hot-swappable PSU has sockets that plug directly into the
power system of the server, which supplies power cables to server components. The power cord does not
attach directly to the PSU (which would require a separate power cord for each PSU). Instead, a single
power inlet serves all power supplies. Many servers offer N+1 expandability for critical components,
particularly the processor and the PSU. “N” is a variable that refers to the quantity of a given component
installed in a system, such as two power supplies. The “+1” refers to a spare component. For example, a
server with three power supplies might be referred to as 2+1, in which two power supplies provide
ongoing power while an additional power supply provides redundancy. N+1 can also refer to a chassis
designed with space to accommodate additional components.

72
Calculating server power requirements
Before calculating power requirements, make sure that the power from the building to the server room is
sufficient to service your equipment. Although most building outlets are 110 volts (V), like household
power outlets, server rooms usually also have 208/220 V outlets to accommodate the high power
demands of larger servers with multiple power supplies and racks full of equipment. As part of your server
room design, make sure the electrical engineers provide plenty of 208 V outlets for present and future
needs. In a rack configuration, a power distribution unit (PDU), similar in function to a household “power
strip” but with much higher capacity, often plugs into the 208 V outlet and supplies power to internal rack
components. Server power supplies automatically detect the voltage of the power source and adjust as
needed. Calculating the power supply needs for a server requires you to know how much power the
motherboard, processor, internal adapters, and peripherals require. The power supply in the server can
probably handle additional components without any problems; however, if you fill all expansion slots and
drive bays, you might exceed power supply ratings. Though some components might list power
requirements on the device or with its documentation, power requirements for other components might
be difficult to locate. In that case, you should err on the safe side and calculate based on the maximum
wattage allowed for a given type of device. The following table is a starting point. When a high-powered
server or rack power is turned on, a sudden, temporary surge of power to the system takes place (known
as inrush power). To account for this, be sure that the electrical engineer not only knows the amperage
requirements for equipment that is up and running, but also accounts for inrush power, usually at least
20 amps (A) per PDU. Otherwise, you won’t be able to power up a rack of equipment all at once without
tripping the breaker.

Some devices might list volts, watts, amps, or combinations thereof. Use the information you gather to
calculate power according to the following formula:
Watts = volts X amps
(This formula involves slight rounding, but should suffice for calculating general power requirements.)

For example, a PCI slot requiring 5 V of 5 A current would require 25W.

Computer components require positive power voltages in +3.3, +5, and +12 V. For example, a hard disk
requires +12 V, and the processor usually requires +3.3 V. You might also find a negative power voltage of
–5 V for backward compatibility with the ISA bus. The motherboard can be designed to supply negative
voltages if the power supply does not. Assume all voltages in this text to be positive unless stated
otherwise.

Add the total power requirements for all server components, and subtract the total number from the
power rating for the power supply. You should have plenty of power to spare, preferably about 6%, if you
want the system to be as reliable as possible. If available power is marginal, you might consider moving
certain components to other servers if possible, or upgrading the power supply. Older computers use a
paddle switch located on the power supply to turn on the computer. More recent servers and
workstations use a remote power switch, which runs cables from the power source connecting leads to
connectors on the switch. Be sure that if you are working inside the case with these wires, you disconnect
the power cable first because the wires carry 110 V AC at all times. Accidentally touching the ends of the
leads together might result in an unpleasant shock.

Uninterruptible power supply

An uninterruptible power supply (UPS) temporarily supplies power using batteries to the load equipment
(anything connected to the UPS that draws power, usually servers and possibly other network equipment)

73
in the event of a power outage. The UPS also supplements power in case of a brownout, where utility
power continues but is below acceptable operating voltages.
A UPS typically provides backup power in that the load equipment constantly receives power from one or
more backup batteries. The batteries receive a constant charge from utility power. If the utility power
fails, the batteries continue to provide power just as they always have, minus the battery-charging
function from utility power. The primary purpose of a UPS is not to continue normal operations for the
entire duration of a brownout or power failure. Instead, the UPS provides a few minutes of power to give
administrators enough time to send network messages to users (giving them time to save and close files)
and gracefully shut down the server using normal procedures in the NOS. Otherwise, users can lose data
from open files and the NOS can become corrupt or unstable.

Power protection systems fall under three major architectures:

• Standby/Offline: A transistor momentarily switches a large transformer, which stores a small
amount of power, before transferring power to the UPS battery.
• These UPS systems are better suited for home PCs, workstations, and so forth.
• Line Interactive: Similar to Standby/Offline, but adds automatic voltage regulation to stabilize
power levels during brownouts and over voltages without using battery power. Used in PC
desktops as well as smaller servers, and usually provides up to 3 kilovolt/amps (kVA).
• Online/Double Conversion: AC utility power enters UPS where the rectifier converts to DC and then
back to AC out to the load equipment. This conversion process cleans the power stream to near
perfection. Used for higher-end, mission-critical servers, and is commonly used where more than
3 kVA is required. (Extremely large power requirements over 3000 VA often require a huge amount
of space, and might be centrally located with specially wired UPS supported building circuits or
power distribution strips.)

Several sources recommend that you do not purchase standby power supply (SPS) equipment, which
detects an interruption in line power and switches to a transformer to bridge the period of time it takes
to switch to battery power. This is no longer an issue, as it was in the 1980s and 1990s, because the quality
of UPS equipment and server power supplies can easily survive the momentary transition without ill
effects. The switch might take 2–4 ms, while power supplies can usually handle a 100–200 ms pause.

Understanding UPS operation

The following description and Exhibit 3-2 describe what takes place during normal operation of an online
UPS when utility power is at normal, uninterrupted levels:
1. The UPS receives AC power.
2. The UPS uses a rectifier to convert AC power to DC power.
3. Some DC power is siphoned off to charge the battery.
4. An inverter converts DC power back to clean, nearly perfect AC power.
5. Power passes through the transformer to the load equipment.

74
 Exhibit 3-2: Normal online operation

The following steps and Exhibit 3-3 describe what takes place during a power failure:
1. Because there is no AC in, the battery discharges DC power to the inverter.
2. The inverter converts DC power to clean, nearly perfect AC power.
3. Power passes through the transformer to the load equipment.

Exhibit 3-3: Battery operation

UPS software
UPS software such as American Power Conversion (APC) Power Chute can also assist the administrator by
automatically shutting down the server and safely storing data that might otherwise be lost. The UPS
connects to a port on the server (usually a serial or USB port) and sends a message to the server when
backup functionality activates. The message serves as a trigger to the software, which begins to perform
administratively predetermined functions such as data backup and system shutdown. Many UPS
manufacturers are including increasingly sophisticated administrative software that goes beyond these
basic functions. For example, Tripp-Lite offers software that allows you to remotely manage UPS systems
of most major manufacturers from an Internet connection.

Failover for the UPS

Backup batteries eventually fail or lose their ability to retain a charge. Server-level UPS systems usually
also offer N+1 functionality that allows you to replace a battery while another battery (or batteries)
continues to power the unit, ensuring that the UPS is not temporarily unavailable. Also, you can use

75
multiple UPS units so that you can service one UPS while the remaining units continue to supply power
(see Exhibit 3-4).

Generally, administrators seek about 15 minutes of backup power for servers.

Depending upon the business need, administrators might seek up to eight hours of backup power, such
as for PBX telecom systems and Internet connections. However, UPS equipment providing eight hours of
backup is large and expensive. If you want backup power for an extended period of time, consider a
backup generator with a UPS.

The primary power service in the event of an outage comes from the generator, not the UPS. However,
the generator usually takes several seconds to come fully online (known as generator kick, kick, or kick
start). The UPS in this context provides power during the seconds required for the generator to come
online.

Exhibit 3-4: N+1 redundancy: if one UPS is unavailable, the remaining units continue to supply power

Determining UPS requirements

Determining your exact UPS requirements involves several variables, including power requirements of
individual servers and networking equipment and/or rack power, one or more monitors, and so forth. The
best way to calculate these needs is to add the total power requirements of the rack or server. Some
sources recommend looking at the UL (Underwriters’ Laboratory) sticker (if present) on the back of the
unit to determine its total power requirement; however, the UL sticker is a measure only of what the
manufacturer submitted as the default configuration for this server, and servers often include additional
equipment. Instead, measure the wattage required by your load equipment. Let’s say that on a particular
server, you need 600 W (which includes 50% overhead, as recommended earlier). When you visit the Web
site of a UPS manufacturer, you only see UPS systems rated in volt-amps (VA), and notice that a volt amp
is calculated using the same formula as watts: volts X amps. Since the formula is the same, does that mean
watts and VA are the same thing? No, they are not. The difference is “where” the power occurs.

76
Power coming “in” from the utility company is measured in watts. That’s what you pay for on your electric
bill, and it is also sometimes called actual power or true power. However, as the electricity passes through
the server’s power supply, capacitors, inductors, and other equipment, we must account for a difference
between the power that comes “in” to the power supply and power that goes “out” of the power supply.
Power going “out” is known as apparent power. The difference between actual power and apparent
power is known as the power factor, which is usually a difference of about 60%.

To boil it all down, it comes to a simple factor in determining UPS power requirements.

Take the VA rating of the UPS and multiply that number by .60 to determine the number of watts that this
UPS will support. For example, if the UPS is 1000 VA, then the watt rating for the same UPS is 600 W (1000
X .60 = 600 W). In the example stated at the beginning of this discussion, a 1000 VA UPS would exactly
meet the 600 W power requirements of the server equipment. However, to plan for future expansion and
add a margin of safety, you should increase to the next higher available VA rating from the UPS
manufacturer.

It is important to realize that adding batteries to the UPS does not increase its volt-amp rating. For
example, adding an extra battery will not upgrade your 1000 VA UPS to 1500 VA. Instead, it only extends
run time. If the run time is 15 minutes, adding an extra battery will power the 1000 VA UPS for a few more
minutes. The only way to increase VA is to use a more powerful UPS.

Next, determine the amount of time (known as the run time) you require powering the server. Realize
that VA X high run time = lots of money. Run time is not a calculation you make on your own; you have to
contact the UPS manufacturer to make that determination. This is best accomplished by visiting a UPS
Web site and using online tools to arrive at the best product for your needs.

Avoid including printers in the total UPS power requirements, because documents can usually be printed
any time; risk of data loss is a more immediate concern. Also, printers can be electrically “noisy,” drawing
varying levels of power; when performing a print operation, they are extremely demanding, electrically
speaking. (Inkjet printers require much less power than laser printers and are better to include on a UPS
if you absolutely must have a printer during a power outage.)

Site preparation
In addition to calculating the total power requirements for your servers, you must also consider the
physical space that a UPS requires. Batteries can be quite large, and several models allow you to daisy
chain multiple external battery packs together to extend UPS run time. This can require a great deal of
space. One solution is a rack-mounted UPS.

All vendors are moving to rack-mounted models in various formats. You can connect UPS units either in
the same rack or in adjacent or back-to-back racks to increase run time or provide UPS fault tolerance.
Also, you can use a rack-mounted UPS with an external battery pack (see Exhibit 3-5). These solutions are
not something administrators arbitrarily piece together. Instead, always consult the UPS manufacturer,
which will advise you on permissible physical connections, safety, power capacity, and run time.

77
 Exhibit 3-5: To provide extra power, connect two UPS systems back to back or use an external battery pack

Rack-mounted UPS systems, especially larger ones, are extremely heavy and usually require assistance to
install and service. Because of the weight, you should place them at or near the bottom of the rack.

The largest UPS systems can be huge. For example, the APC Silicon DP3500E is 70 inches high, 94 inches
wide, 31 inches deep, and weighs a staggering 5500 pounds.

Regardless of the size, the following checklist will help you to plan site requirements, particularly for larger
systems:
• For larger UPS systems, notify building engineers, electricians, and electrical engineers; for
extremely heavy systems, also notify structural engineers. Verify that all aspects of the site are
safe.
• Do not place monitors or other devices that are highly sensitive to electromagnetic fields (EMF)
near the UPS. A larger UPS might emit an EMF that affects computer monitors, but probably not
other equipment.
• UPS systems generate heat. Make sure the HVAC system can accommodate the heat output.
• Plan for any rewiring so that load equipment can reach the UPS. To prevent overloading the circuit,
place larger UPS systems on a single, dedicated circuit.
• Verify sufficient clearance for adequate airflow, probably around 12 inches (0.31 m) behind a
freestanding UPS array and 48 inches (1.22 m) in front (see Exhibit 3-6). Ensure adequate space to
reach all switches, jacks, outlets, and so forth.
• Many UPS units have an optional earth ground to which you can ground the unit (in addition to
existing AC grounding). Locate a suitable ground location reachable from the UPS.
• Get help—even in smaller units, the batteries are surprisingly heavy.
• Some larger freestanding units are on rolling casters to make movement easier.

Make sure you lock the wheels when not moving the unit.

78
 Exhibit 3-6: Allow enough space for airflow and equipment access

The following is a list of major vendors of UPS equipment:

Tripp-Lite—www.tripplite.com
• American Power Conversion (APC)—www.apc.com
• Best Power—www.bestpower.com
• Liebert—www.liebert.com
• MGE—www.mgeups.com
• Oneac—www.oneac.com
• Sutton Designs—www.suttondesigns.com

3.2 THE RACK

In addition to configuring the system itself, you must determine the best physical orientation for the
server—freestanding or rack mounted, and except for the smallest networks, you should plan to install
servers in racks.

Racks provide an advantage in server rooms where the need for server and network equipment grows but
floor space does not. By stacking equipment in a rack, you increase computing assets vertically in the same
floor space that would have otherwise consumed precious floor space. This space savings is known as
density. For example, in Exhibit 3-7, six servers on a table require about six feet of floor space. Using racks
in the same six feet, you can install 18 8-way servers.

79
 Exhibit 3-7: Density allows for more equipment in the same horizontal floor space
If you have existing servers in the tower configuration and want to place them in a rack, you can either
place them on a vented shelf that mounts inside the rack or obtain a vendor kit that adapts the server to
lie on its side and install it in the rack along telescoping rails. Nearly all network equipment (except,
perhaps, for small workgroup hubs or switches) is rack-mountable. When you decide to use a rack, you
must carefully consider several factors, including heat, ventilation, power, weight, grouping, and
accessibility.

Physical characteristics
Physical rack characteristics vary from one vendor to the next. However, general characteristics are listed
here. All aspects of equipment installation revolve around the physical dimensions of the rack as follows
(see Exhibit 3-8):
• Units: Rack equipment is measured in EIA (Electronic Industries Alliance) units, or U. One EIA unit
(1U) is equal to 1.75 vertical inches (4.45 cm). For example, Tripp-Lite makes a rack-mounted UPS
that is only 1U in size. Server appliances are usually between 1U for network appliances dedicated
to a single purpose and up to 7U (12.25 inches, or 31.12 cm) for 8-way servers. Datacenter servers
larger than 8-way (such as a 32-way) are usually in their own dedicated enclosure.
• Height: Including frame, bezels, and feet or rolling casters, a full-height rack (42U) is about 6 feet
(1.8 m). Various manufacturers make smaller racks as well, with common sizes at 22U, 24U, and
36U. Some manufacturers offer rack extensions, which add about 8U to the height. See the
following table as a reference (casters or feet not included).
• Depth: Racks are about 36 inches (.98 m) deep, with usable depth around 28 or 29 inches (between
.71 and .73 m). Try not to use space beyond the usable depth, because you will still need room at
the back to work, and for PDUs, cables, and other specialized devices that you do not need to see
from the front.

80
 Exhibit 3-8: 42U rack dimensions

• Width: Rack-mountable equipment requires a RETMA (Radio Electronics Television Manufacturers

Association) industry standard opening of 19 inches (48.26 cm). Exterior width is usually about 23–
24 inches (58.5 cm).
• Weight: Weight varies depending upon whether the rack includes ballast or doors, which weigh
about 30 pounds each. 42U racks can weigh around 160– 250 pounds empty, and can
accommodate about 2000 pounds of equipment. (IBM has a heavy-duty rack that weighs 575
pounds empty!) 24U racks are about 210 pounds empty, and can accommodate about 1000
pounds of equipment.

Racks can also include several other items:

• Stabilizing feet: Moving rack equipment in and out while servicing it presents a physical danger of
the rack falling over. Stabilizing feet extend beyond the rack and help prevent it from tipping over
when you work on equipment. The HP Rack System/E includes both a front and rear retractable
anti-tip foot for added safety.
• Leveling screws: Similar to the leveling screws under a washing machine (but capable of much
greater weight loads), you turn the screws until the rack is level.
• Wheels: Usually made of polyurethane to reduce the effect of bumps, jolts, and uneven floors
when moving the rack. The wheels can support a great deal of weight, usually 1000 pounds each.
• Filler panels: Purely cosmetic, these panels cover up empty slots for a more professional, finished
look.
• Side panels: Also cosmetic, side panels cover up the exposed sides of the rack.

These are unnecessary where you join two racks together.

• KVM/concentrator/switchbox: A keyboard, video, mouse (KVM), also called a concentrator or
switchbox, enables you to control multiple servers from a single keyboard, video monitor, and
mouse.
• Cable management arm: It is important that when you service equipment and pull it out of the
rack, all the cables do not come loose. Instead of bunching up the cable slack in a tangled mess at
the back, you can Velcro cables to a cable management arm (CMA) that keeps cables neat while

81
 allowing them to extend when you pull out equipment. Cable management brackets also can be
used to guide cables vertically within the rack.
• Ballast: Dead weight placed vertically, at the sides of the rack or at the bottom of the rack, to add
stability when heavy equipment is required higher in the rack. Single ballast usually weighs about
30 pounds.
• Short rear door: The rear door has a gap of a few inches at the bottom to facilitate cabling out of
the rack while maintaining security.

Cooling
Many racks include an option for front and rear lockable doors. Most doors are perforated safety glass or
steel, providing about 60% opening for adequate ventilation, which in most racks is through convection
(see Exhibit 3-9). Some manufacturers (such as Compaq) offer a multi-angled door design that enhances
convection cooling. Warm air rises to the top of the rack, which may also be perforated.

Exhibit 3-9: Perforated doors assist cooling through convection and fan-propelled airflow through equipment

To ensure adequate cooling in the rack, you can install fans in the top panel to increase availability of cool
air inside the rack and draw warm air out (see Exhibit 3-10). Other temperature-control solutions cool
from the front to the back, which is a function of the equipment in the rack. For example, a server might
have a fan at the back of the unit that draws air into louvers at the front, and expels warm air out the
back. Even in this case, you might still install fans at the top, which will help to more quickly expel naturally
rising warm air while also drawing cooler air from the bottom of the rack. (Recall that some floors have
air-conditioning vents beneath the racks.) This is also a more significant concern when the rack includes
items that run hot, such as UPS systems and disk arrays.

82
 Exhibit 3-10: Fans in the top of the rack draw warm air out while cool air enters from the bottom

Be careful not to place anything on the top surface of the rack, which would inhibit dissipation of warm
air. Be careful not to place racks so that the back of a rack faces the front of another to avoid the intake
of warm air. Instead, you should place racks back to back and allow adequate space between rows of racks
(see Exhibit 3-11).

Exhibit 3-11: Arrange rows of racks in a back-to-back configuration

Configuration
Configuring the rack well directly affects the level of practical usability and prevents the need to rearrange
equipment in the future. Major vendor Web sites offer rack configuration utilities as either downloadable
programs or Web-based applets. These utilities are the most effective way to configure a rack, especially
if all rack equipment is from the same vendor. If you make an unwise choice, the utility might alert you
and ask if you want to select an alternative. Even if you do not stay with the same vendor for all equipment
in your rack, you can use a rack configuration utility to approximate similar equipment. For example, the

83
HP Net server LXR8500 is very similar in power requirements and dimensions to the Dell Power Edge 8450.
The following is a short list of vendors offering rack configuration utilities:
• Dell—www.dell.com (download). A good, basic rack configuration utility.
• Compaq—www.compaq.com (Web-based). Offers the advantage of not having to install yet
another program on your computer. Offers two modes, one for novices and one for those more
experienced in rack configuration.
• Hewlett-Packard—www.hp.com (download). This one is excellent in terms of usability and has a
low learning curve.
• IBM—www.ibm.com (download). The most detailed tool, but it will take a little longer to learn its
usage and options.

While extremely useful and educational, these utilities cannot account for every contingency, and you will
have to monitor the results for accuracy and practicality in your own real-world environment. Also, most
of these utilities do not fully consider the implications of joining two racks together, especially in terms of
sharing cabling between the racks. Therefore, you should be able to wisely configure a rack for weight
distribution, device grouping, and cable management as described in the following sections.

Exhibit 3-12: The IBM Netfinity Rack Configurator is an excellent tool

Weight distribution
Larger 8-way rack servers can weigh upward of 175 pounds, and UPS systems like the APC Smart-UPS 5000
RM is 5U and weighs 320 pounds! The general rule of thumb is to place the heaviest items at the bottom
of the rack, which minimizes the chances of the rack tipping over. Therefore, unless the UPS is small (1U
or 2U), you should almost invariably place it at the bottom. If other planning factors such as device
grouping prevent you from placing all the heavier items at the bottom, try to avoid placing heavier items
any higher than 36 inches (.91 m) up the rack, and consider adding one or more ballasts. Each ballast is
1U and weighs 30 pounds. Other heavy items include mass storage items such as a DLT
(Digital Linear Tape, a backup device) and a disk array. (HP recommends placing the DLT just above the
UPS.) It should not adversely affect weight distribution if you need to insert a keyboard/mouse between
heavy devices. (See Exhibit 3-13.)

84
It is very important that you pull out only one piece of rack equipment at a time to avoid tipping. Even
equipment that is only moderately heavy has a greater impact on server balance when it is pulled “out of
center.” A rack can easily exceed 1000 pounds, and could seriously injure or kill someone if it fell over.
When working on a rack, don’t forget to extend stabilizing feet if they are available.

Exhibit 3-13: Pull out only one device at a time, and use stabilizing feet to prevent tipping
Device grouping
Grouping devices involves a delicate balance between weight distribution and logically placing
components where cables from one component can reach other components. Also, for purposes of
usability, some types of equipment should be placed where they are most usable. For example, although
the keyboard is lightweight, placing it at the top of the rack would make it inaccessible to administrators.
If administrators work on the server in a standing position, you should place the keyboard in the middle
of the rack (or a little lower if administrators will be seated). The logic of some devices is also important.
For example, certain devices might need to be closer to one another for cables to reach. Another reason
to group equipment in a different way than you would if only considering weight might be clustering.
Recall that clustering involves two or more servers serving data from the same physical media (hard
drives). In the rack, you might have (starting from the bottom) at least one UPS, Cluster Server 1, a disk
array that contains data, and then Cluster Server 2 (see Exhibit 3-14).

Exhibit 3-14: A cluster might change the usual weight distribution order

Cable management
One of the primary goals in cable management is neatness. While this might sound compulsive,
organization is critical in the server room. When equipment fails, administrators must have minimal
distraction in finding the proper equipment. When trying to access equipment at the back of the rack (to
85
replace a power supply, for example), you do not want to fight your way through a mess of cable. With
cables neatly aligned along a CMA and placed out of the way with cable management brackets, guides,
and so forth, you should be able to quickly access the equipment you need.

Another factor that affects where you place equipment is cable length. The effective cable reach of pieces
using a CMA will be shorter because several inches (perhaps 28 or more) will be used by the arm. However,
the shorter cable reach is well worth the improved organization and secure connections that a CMA
provides (see Exhibit 3-15).

Exhibit 3-15: The CMA expands and contracts when you move the server
If cable lengths are too short, you might have to place dependent equipment closer to one another. For
example, a light, 1U Web server near the top of the rack must back up its content to a DLT located at the
bottom of the rack. If the cable lengths are not sufficient, you might have to move the heavy DLT higher
in the rack, lower the 1U Web server, or move the 1U Web server to an adjacent rack in a lower position
closer to the DLT. Many devices cannot use an extension to make up the shortfall in cable lengths. For
example, there is no such thing as a SCSI cable extension.

Cable lengths become a more visible restriction when you consider a monitor, which is usually placed
higher in the rack than the server. Typically, the monitor cable can reach the server over a maximum of
29U. If necessary, you can purchase an inexpensive VGA extension cable. Use as short a length as
necessary (hopefully 6 feet or less), because longer lengths are not good for video signal integrity.

In the absence of a CMA, you might consider using a straight point-to-point cabling method with no
intermediate cable management. However, this is the least desired method because the rack becomes
more difficult to manage when you have cables hanging in the way. Also, unless there is a method to
secure the cable ends (such as thumb screws), they are more likely to fall off. Some rack equipment has a
cable tray option extending from the back of the equipment. The point of the cable tray is to lessen the
pull of gravity on the cable connection and reduce the likelihood of the cable falling out (see Exhibit 3-16).

Exhibit 3-16: A cable tray helps cables to stay plugged in

Sometimes you need to use multiple devices from several racks that connect to a physical and logical
center. For example, Exhibit 3-17 shows a centrally located server that connects to several devices such
86
as DLT backup equipment, a SCSI disk array, and a Fiber Channel disk array. If other planning factors
prevent you from placing all the equipment in the same rack, then centrally locate the server between
racks containing the equipment.

Exhibit 3-17: Place the server central to the other equipment to which it connects

Power
A UPS is only as good as its availability, so each piece of rack equipment must be able to connect to the
UPS. The UPS has a limited number of outlets (perhaps six), and in a fully loaded rack, that’s probably not
enough. However, by using a rack-mounted PDU, you can create more outlets (see Exhibit 3-18). Make
sure that the devices you attach to it do not exceed 80% of the PDU’s power rating. You can install the
PDU in the rack in any of several configurations. Usually, you try to locate the PDU at the rear of the rack
and as close to the bottom as possible.

Vertically installed PDUs attach to either the left or right rack post (see Exhibit 3-19).

However, most PDUs use 11U vertically, so you can probably only install three on one rail, left or right.
Another factor that limits PDU installation is the type of server. Some servers slide out of the rack at the
rear instead of the front (the HP Net Server LXr Pro8, for example). In this case, a horizontal PDU in the
same space as the server would interfere with server removal. In a vertical PDU configuration, also be
careful not to install the PDU at the location of the locking latch, and face the PDU inward so that rack
covers do not block outlets.

Exhibit 3-18: The PDU provides the functionality of a power strip at much higher power capacity

87
 Exhibit 3-19: A vertical PDU installation

A horizontal installation usually allows for cleaner routing of power cables, and can be mounted behind
most rack equipment in the same EIA unit except for very deep, large rack components such as some large
servers. If installation behind a unit is not acceptable, you can also mount PDUs in the bottom of most
racks (see Exhibit 3-20).

Exhibit 3-20: Mounting in the bottom of the rack does not interfere with other equipment

Many PDUs, especially those with higher voltage, have locking plugs that plug into a locking receptacle to
prevent accidental disconnection. The rack utility software from several vendors calculates the VAs
required to service the equipment you propose, and also suggests the quantity and type of PDUs you will
need. For an added level of power redundancy, you can use a PDU designed for utility power from two
independent circuits. If one of the circuits fails, the other transparently continues to provide service (see
Exhibit 3-21).

88
 Exhibit 3-21: For added redundancy, use a PDU that connects to two independent circuits

The KVM
It is impractical to have a separate keyboard, video display, and mouse for each server in the rack. A 42U
rack can have 42 separate 1U servers! Instead, it makes more sense to install a single keyboard, video
display, and mouse (KVM) that can service all of the connected servers, collectively referred to as a console
(see Exhibit 4-22). A KVM console also reduces air-conditioning costs by eliminating multiple heat-
generating monitors. You can obtain a basic, inexpensive KVM console for any configuration of
workstations or servers, most of which operate between four and eight systems. Some can also be set to
continually cycle between servers every few seconds so that you can observe activity on each server.

Exhibit 3-22: Use one keyboard, video, and mouse to control servers through a KVM console

For the rack, you can use several KVM components to arrive at the configuration you want:
• 1U fixed keyboard/mouse tray
• 2U retractable keyboard and mouse tray
• Full-size (about 11U) video display
• 1U or 2U integrated keyboard, trackball, and video display (see Exhibit 3-23)

89
 Exhibit 3-23: An integrated keyboard, trackball, and pop-up video display

Whichever configuration you decide on, the clear advantage is that you conserve space by using a single
KVM console to manage multiple servers. Also, since each KVM console often manages up to eight servers,
vendors usually offer the option to cascade the consoles, which allows you to manage dozens of servers.
Some consoles are wireless, and with signal amplification they can allow you to control more than 60
servers.

The keyboard, mouse, and video monitor each connect to the KVM console. Then you need one extension
cord (male to female) each for the keyboard, video monitor, and mouse for each server. If you have four
servers, then you need four sets of extension cords. You can usually purchase the set as a bundled pack
from any computer supply source.

KVM switches are available from nearly all major server vendors and most computers stores. APC
manufactures an outstanding KVM console that, in addition to standard functions, includes the following
features:
• Password security to the KVM console adds an extra level of security.
• An on-screen display with customizable menus allows you to graphically switch between servers.
• Hot-pluggable operation allows you to add servers without having to first power off the KVM or
other servers attached to the KVM.
• Mouse reset circumvents a frustration with KVM systems in which the mouse ceases to respond
to actions. This feature also allows you to regain control of the mouse without powering down the
KVM or the server.

Alternatively, you can usually administer servers from the comfort of your own desktop with a fullscreen
monitor, keyboard, and mouse, depending on the operating system.

For example, you can use the text-based Telnet utility to remotely connect to and administer UNIX systems
and perform router programming. Windows 2000 and NT 4.0 offer administrative tools that you can run
from any Windows computer on the network.

NetWare offers similar management control.

Blade servers
Blade servers are large cards containing the essential elements of a server – processors, memory, and
hard disks – that slide into a chassis alongside like servers, like books on a shelf.
90
The blade server chassis, which can typically hold six to twelve or more blades, provides redundant power
supplies, fans, network connections, KVM switches, and connections for external storage. The chassis
mounts in a regular rack and reduces the number of needed cables. In addition to easy deployment and
replacement of servers, management software allows for centralized management of individual servers
or whole groups of servers.

3.3 RACK INSTALLATION TIPS

The following tips can help you to properly configure the rack:
• Begin building the rack from the bottom up, with heaviest devices at the bottom (as previously
discussed).
• If using multiple racks, be consistent in the server numbering scheme in relation to port numbers
on the KVM, which helps you to quickly access the right server from one rack to the next instead
of using a hit-or-miss method.
• Place servers attached to the same KVM console in close proximity to it so that you can easily view
server activity while at the keyboard.
• Group servers that serve a particular purpose in the same physical area. For example, put Web
servers in one location, file servers in another, application servers in another, and so forth. This
helps to minimize trips across the server room when configuring or troubleshooting.
• Remember that at times you must access the rear of the server. Make sure there is adequate space
behind the server to open the rear door.
• Place servers that require frequent access (such as those with backup devices) in more easily
accessible locations.
• Place servers according to security need. Position high-security servers in highly visible locations
where it is more obvious if someone is accessing them.
• For added security, lock the rack doors and install an alarm that trips when the door has been
breached.
• Fully loaded racks are extremely heavy, so limit the number of racks that you tie together if you
plan to move them, even for routine cleaning. It can be very difficult to roll three attached racks,
even with assistance.
• When sliding equipment in and out of a rack, watch your fingers! The slide rails are pinch points.

Hundreds of pages about rack planning and installation can be found at the following sites:
• http://netserver.hp.com/netserver/docs/download.asp?file=g_rack_cabling.pdf
• http://support.dell.com/docs/systems/smarcon/en/index.htm
• ftp://ftp.compaq.com/pub/products/storageworks/techdoc/racksandoptions/ 14255e2.pdf
• www.ibm.com (search for the Netfinity Rack Configurator)

Review Questions

1. Server rooms usually require what voltage outlets?

2. In a rack configuration, a PDU functions as what common household item?
3. What is the formula for calculating server power requirements?
4. What is considered load equipment?
5. What is the purpose of a UPS?
6. What good is N+1 redundancy with a UPS?

91
7. What is the simplest way in determining UPS power requirements?
8. What is a primary advantage of a rack?
9. How big is 1U?
10. Where should you generally place the heaviest items in the rack?
11. What helps to manage cables in the rack?
12. Where should you usually mount a PDU in the rack?
13. What is a KVM console?
14. What is the advantage of using a
15. KVM console?

92
 TOPIC 4
_____________________________________________________________

4. HARD DISK INTERFACES AND RAID

_____________________________________________________________

LEARNING OUTCOMES
After studying this topic you should be able to:
• Identify basic physical hard disk components, compare physical and logical drives and describe
their functionality, and identify major file systems.
• Identify characteristics of the IDE interface and configure IDE cabling and connectors.
• Identify characteristics of the SCSI interface.
• Configure SCSI cabling and connectors.
• Become familiar with Fiber Channel technology and storage area networking and identify and
configure various types of RAID.

4.1 HARD DISKS

The physical hard disk is the focal point of the enterprise. The operating system, applications, and data
are stored on the hard disk, and without hard disk storage, there is nothing for anyone to “do” on the
network. Therefore, it makes sense that server storage be protected against failure and that
administrators configure storage so that it is as fast as reasonably possible. This unit opens with a brief
inventory of hard disk physical components, differentiates physical and logical drives, and describes the
major file systems commonly in use on servers. To operate the hard disk, an IDE or SCSI controller and
attachment interface are necessary. Also, you will need to be aware of rules that govern appropriate
connections and compatibilities for a given hard disk configuration, especially for SCSI.

Although desktop users commonly use only a single physical hard disk, server administrators often
manage dozens of hard disks—most of which require high performance, redundancy, or both using
various RAID implementations. When administrators add or replace a drive, it is vital to do so with minimal
disruption (if any) to users, and this chapter shows you various ways to do this. Part of administering
server storage is proper maintenance, and you will learn about several such utilities. Finally, you will learn
about network storage, including a new technology known as Fiber Channel that provides outstanding
performance.

Hard disk components

Although we use the term “hard drive” or “hard disk,” a hard disk is really multiple physical platters inside
a sealed, dust-free housing. The manufacturing process is extremely controlled—there is no tolerance for
contaminants (such as dust particles) — and manufacturing specifications are extremely tight. Most hard
disks include the following major physical components (use Exhibit 4-1 for reference as you read the rest
of this section):
• Disk platters
• Drive heads
• Actuator mechanism

93
 Exhibit 4-1: Hard disk components

The hard disk also consists of the following components not visible in Exhibit 4-1:
• Spindle motor
• Logic board
• Connectors
• Jumpers

Though it would be impractical to exhaustively describe each hard disk component, you should be aware
of their basic functionality.

Disk platters
The disk platter is a rigid disk inside the sealed hard disk enclosure. In the past, disk platters have been
composed of a metallic aluminum/magnesium alloy, a lightweight and rigid material. On the surface, syrup
containing iron-oxide particles was evenly spread across the disk using centrifugal force. It is this material
that stores the magnetic data on the platter. With the cover off the sealed enclosure, you can see that the
platter is a brownish or amber color. While you might still have disks in use that were manufactured using
this process, it is not implemented in current manufacturing because the oxide medium is soft, making it
susceptible to damage if the drive head touches the surface during operation, usually due to a jolt or
vibration. (This is known as a head crash.) A head crash often (but not always) corrupts the data or
compromises the integrity of the recording media.

Current manufacturing procedures have abandoned the metal platter in favor of a glass platter (actually
a glass-ceramic composite) because it does not flex as metal can, allowing the platter to be about onehalf
the thickness of a metallic platter. Thinner platters allow room for more platters in the same drive housing,
hence higher-capacity drives. Most current hard disks utilize four platters, but high-capacity drives can
use upward of 11. Also, the glass platter is more thermally stable than metal, minimizing the expansion or
contraction that occurs with temperature changes.

One way to apply the medium to the platter is to process the platter through a series of chemical baths
that leave several layers of metallic film in an electroplating method similar to that used to affix chrome to
car bumpers. A better, newer method is to apply a thin-film magnetic medium over the glass platter,
providing greater density. The thin-film medium is more expensive than electroplating, and it is applied
through a process known as “sputtering” in which the material is applied in a continuous, nearly perfect
vacuum. Thin-film media are much harder than oxide media. The result is that it is more difficult (nearly
impossible) to crash the media. Thin-film media are like a silvery mirror in appearance.

94
Drive heads
The drive head reads and writes data to the magnetic media on the disk platter. Each platter has two drive
heads: one that reads media from the top and another that reads media from the bottom. When
stationary, the drive heads are actually parked on the media surface. When the drive is in use and spinning,
the air pressure from the movement of the platter separates the drive heads from the surface of the
media. Because the drive heads are ganged together in a head rack to a single actuator mechanism, each
head moves across its respective platter in unison with the other heads of the drive.

Older hard disks required a program to manually park the heads when powered off to avoid having the
heads skitter across the surface, causing damage. Today, drives automatically park their heads using a
spring and magnetic force. When the drive powers off, the magnetic force of the voice coil actuator
dissipates, and a spring drags the head rack to a park-and-lock position.

The actuator mechanism is the mechanical component that physically positions the drive heads at the
appropriate location on the disk platter to read or write data. Most actuator mechanisms today use the
voice coil construction, which derives its name from audio speaker technology using an electrically
charged coil. Fluctuations in the electrical charge move the coil to various positions over the platter similar
to the way a speaker coil moves to create audible vibrations in a speaker cone. The actuator mechanism
has no intelligence of its own in determining the appropriate location; it depends upon the servo
mechanism, which detects precise cylinder locations on the platter using gray code, a special binary code
written to the drive by the drive manufacturer that identifies physical locations on the drive. You cannot
alter or erase this code, even with the FDISK utility or FORMAT command.

The physical components described above are relatively durable considering their precise nature. For
example, specifications for many drives indicate the MTBF is approximately 1,200,000 hours (that’s almost
137 years) and can sustain shock up to 300 Gs (the force of gravity times 300).

The environment is still extremely important for hard drives, particularly the temperature. We have
already discussed adequate ventilation and cooling, but if your office is in a colder environment, also
consider condensation. When a drive comes delivered to you from a cold truck in the middle of winter,
allow plenty of time for the drive to warm up (acclimate) to room temperature before using the drive to
prevent condensation on internal components. If the drive has been in an environment colder than about
50 degrees F (10 degrees C), allow it to sit at room temperature for several hours before opening the
package.

Partitions and logical drives

Nearly all operating systems use a storage system that begins with basic Microsoft MS-DOS hard disk
partitions. During OS installation, many operating systems (NetWare, for example) might modify the
partition or create another partition. To further subdivide and organize storage on partitions, you create
logical drives. This section briefly discusses partitions and logical drives.

Creating a partition
The operating system—whether it is as simple as MS-DOS or as sophisticated as NetWare—requires a
defined boundary on the hard disk on which to place its files. The purpose of the partition is to provide
this boundary. A partition can be a primary partition or an extended partition (see the section on logical
drives). A primary partition is a bootable partition on which you can install operating system files. The MS-
DOS FDISK utility is usually used to create the partition. You can create up to four primary partitions using
Windows NT/2000 disk management, but this will be unusual unless you intend to install multiple
95
operating systems on the same server (as in a classroom or lab environment). Typically, you create one
primary partition and an extended partition, which can, in turn, contain logical drives.

After creating the partition(s), you must reboot before using the FORMAT command to create the file
system. If you try to format without rebooting, you will get an error message.

Even if you don’t use Windows 98 in your environment, obtain a Windows 98 startup disk because it has
several useful utilities, including FDISK and FORMAT. Booting from the Windows 98 startup disk
automatically loads drivers that work with most CDROMs, avoiding the need for you to create a
customized disk with your specific CDROM drivers. Also add the SMARTDRV.EXE utility and load it prior to
installing the NOS because its ability to cache file reads in advance significantly decreases installation time.

Run the FDISK utility by typing FDISK at the MS-DOS command prompt. With more recent versions of FDISK
(as in Windows 95B or later), notification that the hard disk is larger than 512 MB appears, and a lengthy
prompt asks if large disk support is desired over 2 GB. Nearly always, you want to respond Yes to this
option. The series of FDISK menu options is easy to navigate

There are third-party alternatives to FDISK and FORMAT. One alternative is GDISK from Symantec, a
command line utility that works much faster and offers more flexibility than FDISK and FORMAT. GDISK
accompanies Symantec Ghost, a disk duplication software product. Another alternative is PartitionMagic
from PowerQuest, which has a more intuitive graphical interface and is a favorite among administrators
for quickly resizing and moving partitions as well as converting them from one file system to another.

Logical drives
A logical drive is a section on the hard disk that appears to the operating system as if it were a separate,
distinct hard disk and has its own drive letter. A logical drive requires an extended partition, which can
take the place of one of the four primary partitions.

You can have a maximum of either four primary partitions or three primary partitions and one extended
partition. The sole purpose of the extended partition is to store logical drive(s). Logical drives can be
lettered up to Z. Within available disk space limitations, you can create as many logical drives in the
extended partition as you want. Use FDISK to create the logical drives.

Once the partitions and/or logical drives are created, you must reboot the system and then format the
drives. If you do not reboot the system prior to formatting, you are likely to receive an error message.
Because the partition is only a storage boundary, you also need to format the partition or logical drive
before you can store files on it. Many operating systems include utilities that allow you to create and
manage additional logical or primary partitions. For example, Windows 2000 Server includes the Disk
Management console with which you can create additional partitions and software RAID configurations.

4.2 FILE SYSTEMS

A file system is a structure that an operating system uses to name, store, and organize files on a disk. As
you administer various operating systems, it is important to understand the basics of each file system.
(The following are general descriptions, and are not intended to be exhaustive.)

96
FAT/FAT32
The Microsoft-based File Allocation Table (FAT) file system is compatible with nearly any operating system
and uses an invisible table near the top of the partition that points to the physical location of data on the
disk. It is the simplest file system, and the network operating systems discussed in this book can all be
installed with a FAT file system, though you will often choose to convert to another file system during or
after the installation, primarily because of the following FAT limitations:
• Small volume size: FAT only supports volumes up to 2 GB in size—tiny even by current home user
standards.
• Large cluster size: The file system stores data on the drive in 32 KB “chunks” known as clusters or
allocation units. If you save a file that is only 2 KB in size, it must use all 32 KB of the cluster, wasting
the remaining 30 KB. The space of wasted kilobytes is known as slack. Even with hard disk storage
at a relatively inexpensive level in recent years, the excessive slack of the FAT file system is
undesirable and quickly adds up to several megabytes of wasted space.
• Limited file size: The maximum file size is 2 GB. While this seems large, it is woefully insufficient
for most corporate databases.
• Security: FAT offers no local security. Therefore, any passerby with a boot floppy can fully access
the files on the local hard disk. Of course, physical security measures should prevent local access;
nevertheless, administrators usually prohibit a strictly FAT file system on the server.

FAT32 is the next (and last) generation of the FAT files system. It overcomes the first two weaknesses of
the FAT file system by offering large disk support up to a theoretical 2 TB and using only 4 KB cluster sizes
for a significant reduction of slack. However, Windows 2000 only allows up to 32 GB. Also, the maximum
file size is 4 GB, but that’s still too small for many corporate requirements. Administrators avoid FAT32
because it does not offer security. Microsoft offers a secure file system with the NTFS file system.

NTFS
The NT File System (NTFS) is the Microsoft Windows NT file system. NTFS is compatible only with Windows
NT 3.1 or later, including Windows 2000/2003 and XP, and is not directly compatible with other operating
systems. An NT/2000/2003/XP machine can have both FAT and NTFS volumes at the same time. NTFS
volumes offer the following benefits:
• Large volume size: NTFS can support extremely large volumes, though the practical limit according
to Microsoft is 2 TB.
• Small cluster size: NTFS formats clusters at 4 KB each by default on partitions larger than 2 GB,
though you can select a cluster size from 512 bytes to 64 KB.
• Large file size: File size is limited only by the available drive space.
• Security: Unless using hacker’s tools, passersby cannot boot and access files on a local NTFS volume
unless they also have a user account that is authorized to access those files. Also, administrators
can apply very specific levels of file and folder security that are unavailable on a FAT file system.
For example, you could allow a user to save a file or folder but not delete files or folders.
• Compression: Though FAT offers compression through Microsoft utilities, these utilities are not
considered reliable enough to be practical for server use. NTFS allows you to compress files and
folders using a highly reliable compression scheme, conserving disk space.
• Data integrity: NTFS includes mechanisms designed to ensure that data is properly and completely
written to the drive.
• Windows 2000 NTFS features: Under Windows 2000/2003/XP, a variety of new features appear in
NTFS 5. You can encrypt files and folders with a nearly unbreakable encryption scheme, and

97
 set quotas so that users do not abuse file storage privileges. Besides the file system itself, the
operating system offers many useful capabilities for managing partitions and files, including the
Distributed File System (Dfs) to deploy what appears to be a single directory structure over multiple
physical file servers. Windows 2000 also has offline storage to migrate seldom used files to a
slower and less expensive storage medium such as tape.

HPFS
IBM uses the High Performance File System (HPFS), which bears many similarities to NTFS in basic
structure. In fact, Windows NT was originally supposed to be named OS/2 Version 3.0 in 1993. However,
IBM is not currently pursuing further developments to HPFS, and so NTFS has surpassed HPFS in many
respects. HPFS is designed to provide local security similar to NTFS, but also coexists with an MS-DOS
partition, allowing you to dual-boot between either MS-DOS or OS/2. OS/2 includes the Journaled File
System (JFS), which contains its own backup and recovery capability. Using an indexing system and log to
corroborate file changes, JFS can interoperate with the operating system to repair corrupt files. HPFS
supports a maximum partition size of 2 TB and 2 GB file sizes, which was a staggering size when OS/2 was
at its peak. HPFS also offered long file name support (up to 254 characters), 512-byte clusters, intelligent
link tracking (you can move a file and the “shortcut” automatically points to the new location), and
efficient file distribution on the hard drive to minimize fragmentation. Along with OS/2, Windows NT 3.x
can also access HPFS partitions.

Linux/UNIX file systems

There are dozens of different “flavors” (versions) of UNIX in use. For the most part, this text and the
CompTIA Server+ exam gravitate toward Linux as the specific implementation of UNIX. UNIX
implementations generally use the UNIX File System (UFS), the Network File System (NFS), or AFS, which
stems from Carnegie-Mellon’s Andrew File System. Linux often uses the Filesystem Hierarchy Standard
(FHS), which is more a directory structure than a file system (see www.pathname.com/fhs for the
complete standard). Finally, you can also create Linux-specific partitions—Linux swap, Linux native, and
Linux RAID.

UNIX file systems have a higher administrative learning curve—you must use an arcane command line
interface. However, several interfaces (such as X Windows) now allow you to perform many UNIX
administrative functions using a GUI. Also, various flavors of Linux characteristically include a GUI.

NetWare
The traditional NetWare file system competes directly with Microsoft’s NTFS, and therefore offers similar
features: You can use very large volumes and files, and the cluster size is efficient. Using the traditional
NetWare file system, you can create a NetWare volume, which is a collection of files, directories,
subdirectories, and even partitions. You can combine separate partitions that together comprise a
volume. File storage is also efficient because NetWare volumes can use sub allocation. A NetWare volume
can subdivide a block (a Microsoft cluster is a NetWare “block”) to minimize slack space. Unfilled blocks
are subdivided into 512-byte sub allocation blocks, which can then be used to store data from one or more
other files.

Like NTFS, NetWare offers file compression. You can manually activate compression by flagging files and
directories with the IC (immediate compress) command or by using the SET command at the server
console to configure a file inactivity delay, after which compression occurs automatically. NetWare offers
a distributed file system and offline file storage similar to Windows 2000 Dfs and offline file storage,
respectively. NetWare 5.x also includes additional features in its optional Novell Storage Service (NSS) file
98
system. The purpose of NSS is to increase performance and total storage capacity. NSS offers several
improvements over the traditional NetWare file system:
• Large files: Instead of the traditional 2 GB limitation, files can now be up to 8 TB each! Also, you
can store trillions of files in a single directory, compared to the 16 million entries per volume under
traditional limitations.
• Performance: Large files typically take a long time to open, but NSS provides rapid access
regardless of size. Also, mounting a volume (preparing it for use) is much faster.
• Flexible storage management: You can create up to eight NetWare partitions on a single disk and
create unlimited volumes per partition.

Note: you can’t just install an NSS file system; you have to first have the traditional NetWare file system
and then add NSS.

The IDE interface

Integrated Drive Electronics (IDE) can refer to any hard disk with an integrated controller. However, it is
more technically accurate to apply the term ATA (AT Attachment) to what we usually call an IDE drive,
because the drive plugs into the 16- bit ISA bus known as the AT bus. In spite of this technicality, this text
and most references use the terms IDE and ATA synonymously. Recall that an IDE drive includes the
controller in the circuitry attached to the drive. When people colloquially say that they are installing an
“IDE controller,” what they really mean is that they are installing a host adapter, the more accurate term
for what is usually referred to as an IDE or SCSI hard disk controller. The host adapter is the physical
interface between the hard disk and the computer bus. As you can see, disk interface terminologies are
replete with inaccuracies (and we have only gotten started). As we delve into the IDE interface and its
variations, you’ll learn synonyms for various interfaces. Other devices besides hard disks can plug into the
ATA interface, usually CD-ROM drives and tape drives. These devices require a variation on the ATA
specification known as the ATAPI (ATA Packet Interface) specification. The SCSI interface, can also accept
other devices.

Early generations of the ATA interface will be discussed in less detail, because you are not likely to find
them in servers. However, some information about them is relevant because standards introduced with
them carry through to current ATA specifications.

ATA-1
The following major features characterize the ATA-1 standard:
• Signal timing for DMA and Programmed I/O (PIO), which utilizes the processor to handle disk
transfers but is super ceded by DMA and Ultra-DMA
• 40/44-pin cable connections (44-pin connections use four more pins to supply power to notebook
hard drives)
• Determination of master, slave, or cable select using jumpers
• Transfer rate of 3.3 MBps to 8.3 MBps depending on the PIO or DMA mode

ATA-2
The following major features characterize the ATA-2 standard:
• Large drive support for up to 137.4 GB (previously 8.4 GB)
• Faster PIO and DMA transfer specifications
• Power management support
• Removable device support
99
 • PCMCIA (PC card) support
• Reports drive characteristics to software (useful for Plug and Play)
• Transfer rate of 8.3 MBps to 16.6 MBps depending upon the PIO or DMA mode

5–12 48573: Systems Support: Assembling, Repairing and Maintaining a Server

Before proceeding, let’s sort out some more terms. ATA-2 is synonymous with the unofficial marketing
terms Fast-ATA, Fast-ATA-2, and Enhanced IDE (EIDE). EIDE has become one of the most accepted terms,
and it applies in a general way to ATA-2 or better. The true published specification is AT Attachment
Interface with Extensions, but people seldom use that title.

In order to support drives larger than 8.4 GB, you must also have an operating system that is capable of
recognizing larger drives. The following operating systems support drives larger than 8.4 GB:
• Windows 95B or later—Using FAT32, you can format drives up to 2 TB.
• Windows NT 4.0—This OS supports larger drives out of the box, but not on the bootable drive.
For the bootable drive, you must first apply Service Pack 4 to enable large drive support.
• Windows 2000—This OS provides native support for larger drives.
• OS/2 Warp—With a Device Driver Pack upgrade, a boot partition can be only as large as 8.4 GB. If
you use HPFS, OS/2 supports up to 64 GB.
• NetWare 5.x—This OS provides native support for larger drives.

ATA-3
The following major features characterize the ATA-3 standard:
• Includes Self-Monitoring and Reporting Technology (S.M.A.R.T.), a predictive technology that
enables the operating system to warn of a device’s degradation. S.M.A.R.T. has its basis in
preceding technologies known as Predictive Failure Analysis (IBM) and IntelliSafe (Compaq). Drives
might use one of the three technologies, or a combination. For example, some Compaq systems
use both IntelliSafe and S.M.A.R.T. If the hard disk begins to show signs of failure, you will see
messages in the server’s system log, RAID log, or in a vendor-supplied monitoring and reporting
utility.
• Optional security mode that protects access to the drive with a password
• Transfer rate of 11.1 MBps to 16.6 MBps depending on the PIO or DMA mode

S.M.A.R.T. has made its way into the SCSI world of hard disks as well. This significantly adds to the
administrator’s ability to monitor the health of internal and external SCSI disks and RAID configurations.

ATA-4
The following major features characterize the ATA-4 standard:
• Addition of the ATAPI standard to attach other types of devices
• Advanced power management
• Specification of an optional 80-conductor, 40-pin cable-select cable to reduce noise
• Improved BIOS support for a theoretical capability of 9.4 trillion gigabytes, though the actual ATA
standard is still limited to 137.4 GB
• Ultra-DMA (UDMA) support, increasing the transfer rate to 33 MBps

100
ATA-4 is probably the most groundbreaking standard in terms of current IDE performance. It introduced
a level of performance (which was increased even further with ATA-5 and ATA-6) that had been formerly
available only on the SCSI interface.
Other terms for ATA-4 include Ultra-DMA and Ultra-ATA. In reference to the transfer rate, you might also
see UDMA/33 or Ultra-ATA/33. In previous ATA implementations, data is transferred once each clock
cycle. Ultra-ATA differs in that data transfers twice for each clock cycle, once at the rising edge and once
at the trailing edge. Ultra-DMA also adds a cyclical redundancy check (CRC) to ensure the integrity of data.

To support Ultra-ATA/33 and later, a compatible drive, BIOS, operating system, and host adapter interface
must be in use. In the BIOS, most manufacturers now include an artificial “32-bit” transfer. Recall that IDE
operates at 16 bits on the ISA interface, even though the host adapter is usually a PCI card. The BIOS now
includes functionality that allows for two 16-bit transfers to occur at once, hence the “32-bit” transfer.

Under ATA-4 and higher, a single drive on the IDE cable must be at the end of the cable (no “stub” allowed).
Otherwise, signaling problems can occur. Under earlier ATA versions, a stub was OK but inadvisable.

If you have any older ATA-1 through ATA-3 drives, don’t throw them away just because current standards
are ATA-4 or better. The ATA specification requires successive ATA iterations to be backward compatible.
For example, you could still attach an ATA-1 drive to an ATA-4 or higher host adapter for cheap (but
slower) storage. Note that as a general rule, if you mix ATA standards on the same cable, both devices
operate at the performance level of the slower standard.

ATA-5
The following major features characterize the ATA-5 standard:
• 80-conductor cable required (as opposed to optional) in order to achieve the maximum transfer
rate. You can use a standard 40-pin cable but only at a maximum transfer rate of 33 MBps.
• Added to the ATA-5 specification is an IEEE-1394 (FireWire) link that allows use of an ATA drive on
the FireWire interface. (FireWire is an extremely fast bus allowing up to 63 connected devices and
up to 3200 Mbps throughput in the latest version.)
• Transfer rate of 66 MBps, achieved by reducing setup times and increasing the clock rate. Later
implementations of ATA-5 achieve 100 MBps under the marketing title of Ultra-DMA/100. This
came about as a result of manufacturers that could match the 100 MBps transfer rate of the ATA-
6 standard but did not want to wait for the completed standard. Starting with ATA-5 and ATA-6,
you are most likely to have to add an adapter to achieve the maximum throughput because it takes
several months and sometimes over a year for motherboard IDE interfaces to catch up to the latest
ATA standard.

ATA-6
This standard includes all features of previous ATA standards plus a formalized 100 and 133 MBps
specification.

SATA
The next step in IDE/ATA technology is Serial ATA, or SATA. Rather than using the usual parallel
communication with a 40-pin IDE ribbon cable, SATA uses point-to-point serial communication. This
requires much smaller cables and connectors with four pins – one pair for transmit and one for receive.
First examples of SATA drives have transfer rates of 150 MBps. SATA II doubles the transfer rate and adds
other enhancements.

101
ATA cable
The standard ATA cable connecting the drive to the host adapter is a 40-pin ribbon cable (see Exhibit 42).
To prevent incorrect connections, a cable key (protruding notch) on the cable matches a corresponding
gap in the IDE connection on the hard drive. The cable should have one striped wire (usually red,
sometimes blue) that indicates Pin 1. Pin 20 is not used and is usually absent from the drive, and a
corresponding block in position 20 appears on the cable, also preventing backward insertion (see Exhibit
4-2 and Exhibit 4-3). Orient the connection so that Pin 1 on the cable is adjacent to the power cable
connection. Because the cable is not shielded, you are limited to a length of 18 inches (457.2 mm). A
longer cable could be more sensitive to timing and electrical noise issues, resulting in data corruption. The
host adapter connected to the other end should also have a cable key, stripe, and absent Pin 20, and most
manufacturers include a marking on the PCB (printed circuit board) that indicates Pin 1.

Exhibit 4-2: Standard 40-pin IDE connector

The cable length of 18 inches (457.2 mm) is usually plenty for a desktop workstation, but in full-size server
tower cases in which the controller is farther away from the drive bays, this might cause a problem. If the
ATA host adapter is a card (as opposed to integrated on the motherboard), move it as close as possible to
the drive locations. Otherwise, if you need longer lengths, you can obtain a longer, custom-made IDE
cable, but at the risk of performance degradation and data loss.

Exhibit 4-3: Correct orientation of the IDE cable on the drive

You can connect a maximum of two devices to a single ATA connector. The cable and its dimensions appear
in Exhibit 4-4. Also, 80-conductor cables are identical except for an additional 40 grounded conductors. At
one end of the cable, an IDE port connector is blue. Two more connectors appear on the cable—the first

102
one is at the opposite end and is black and the middle one is gray. Connect the first hard disk to the black
end connector. If you only have one hard drive, also make sure you connect it to this one and not the
middle connector. Otherwise, the dangling connector on the end (the “stub”) is not well terminated.
Although it is not against ATA specifications, attaching a single drive to the middle connector is not
recommended. Connect a second hard drive to the middle gray connector.

Exhibit 4-4: ATA 40-pin color-coded cable

As an option, you can use the 80-conductor 40-pin cable with ATA-4, but it is required with ATA-5 and
ATA-6 (see Exhibit 4-5). An 80-conductor 40-pin cable begs the question: If there are 80 conductors, why
isn’t it an 80-pin cable? The additional 40 conductors are connected to ground only, and do not have a
corresponding pin on the drive. Remember that an IDE cable is not shielded in any way, and is susceptible
to electrical noise, usually crosstalk from adjacent conductors (hence the ground wires between signal
wires). The extra grounded conductors absorb much of the electrical noise that would otherwise defeat
the added performance of ATA-4 and better drives.

Exhibit 4-5: An 80-conductor, 40-pin cable

If you connect a faster device such as a hard disk and a slower device such as a CDROM to the same IDE
cable, the hard disk performance will suffer while waiting for the CD-ROM to finish its tasks. Because IDE
is not capable of simultaneous I/O, one device must wait for the other before performing tasks. In this
case, it would be better to obtain another ATA host adapter and keep slower devices (such as CD-ROMs
and tape drives) on one IDE connector and hard drives on the other.

Master, slave, and cable select

Because you can place two drives on an IDE host adapter, there must be a determination as to which one
is the master and which one is the slave, especially because the master receives a drive letter assignment
from the operating system first, and is also the device on which a boot record must be found. Otherwise,
the master and slave drives are equivalent despite the implication of the names. You can use two methods
to specify which drive is master and which is slave: First, you can set the jumper on the drive to indicate a
master or slave setting. The drive manufacturer specifies these jumper settings in the drive manual or

103
online documents. For example, Exhibit 4-6 shows the settings downloaded from Maxtor for the Diamond
max 80 Ultra-DMA/100. Most manufacturers also include an indication on the PCB or somewhere on the
drive case as to which jumpers to set for master, slave, or cable select. Although you might need to change
the jumper setting to specify master or slave for some older IDE drives, newer drives allow cable select,
which means that the drive’s position on the cable indicates whether it is a master or slave. Place the drive
on the end of the 80-conductor cable to make it a master, and on the middle connector to make it a slave.

Exhibit 4-6: Jumper settings for master, slave, or cable select

It’s not as much an issue as it used to be, but sometimes if you mix two drives from different vendors on
the same IDE channel, you may experience problems. For example, the system might be unable to detect
one of the drives, or if the drives are different sizes, report both drives at the smaller size. If you experience
problems of this kind and are certain jumpers and locations on the IDE cable are correct, consider placing
drives of the same manufacturer (and perhaps the exact same model) on the same cable.

The pros and cons of IDE/ATA

Current ATA implementations have some compelling advantages for use in the server world, but also have
several disadvantages. If the disadvantages outweigh the advantages, and budgeting is sufficient, you will
probably choose the more expensive SCSI interface.

Pros
• Inexpensive—both the drives and host adapters are very inexpensive; even administrators on the
tightest of budgets can probably afford a few new IDE drives.
• Reasonable performance—in less demanding situations where the server is not a heavily utilized
file server, the reasonably good performance might be acceptable considering the low cost.
• Simple configuration—Plug it in and format the drive. That’s pretty much it, and you don’t usually
have configuration complications, which you might experience with SCSI, for example.

Cons
• Only two drives—Two drives might seem like a lot for a workstation, but a server often connects
to many more drives for purposes of performance, redundancy, and total available storage.

104
 • Slower device throttling—mixing slower devices with faster ones can drag the performance of the
faster one.
• Cable length—18 inches (457.2 mm) is very limiting in a full-size server.
• No simultaneous I/O—only one IDE device can operate at a time on a single IDE channel. SCSI,
which does permit simultaneous I/O, might be much more attractive in terms of performance.
• No native redundancy—you cannot take a standard ATA drive and host adapter and configure
hardware RAID. However, software RAID is available on certain operating systems (Windows 2000,
for example). Also, some manufacturers such as Promise Technology Inc. (www.promise.com) are
making ATA adapters that allow you to configure several levels of RAID.
• Despite advertised transfer rates, most of the time you only get that rate from cached data. Data
retrieved fresh from an Ultra-ATA/100 hard disk can usually only is sustained at about 40 MBps.

Although simultaneous I/O is not supported on a single IDE channel, you can place two drives on separate
IDE channels. In this configuration, each drive can operate simultaneously. (More information about
ATA/IDE can be found at www.ata-atapi.com and especially www.pcguide.com.)

4.3 THE SCSI INTERFACE

SCSI (Small Computer System Interface) technology, like ATA, has seen multiple generations of
specifications, and includes an abundance of technical details beyond the scope of this book. However, as
a server administrator, you must be aware of the major SCSI standards in order to provide the best
available compatibility solution with the best possible performance. Properly configuring your SCSI hard
drives also avoids complex troubleshooting.

As with IDE, SCSI drives include the controller circuitry directly on the hard disk assembly (HDA). In fact,
some sources claim that hard disk manufacturers nearly mirror their drives in all physical respects and
only change the circuitry to distinguish IDE and SCSI, often differentiated by only a single chip. What most
people call the SCSI “controller” is actually the SCSI host adapter—similar to the inaccurate reference to
an IDE “controller.” However, although an IDE device communicates directly on the system bus, a SCSI
device first communicates with the CPU through the SCSI host adapter.

At the end of this section, you will see a listing of the various pros and cons of using the SCSI interface.
Generally, however, the main “pro” is that it performs exceptionally well in multiple disk configurations.
The main “con” is that when you combine the more expensive SCSI hard disks with the more expensive
SCSI host adapters, the price tag is significantly higher than an IDE configuration. SCSI standards range
from SCSI-1 to SCSI-3 and encompass several variations in between.

You are unlikely to find SCSI-1 in modern server implementations, but you need to know some basics
about it according to the stated CompTIA Server+ exam objectives. SCSI-2 (especially the Fast and Wide
iterations) and some SCSI-3 variants are more likely to be found in servers today. With most SCSI
standards, you will see which cables and/or terminators are required, all of which are explained in more
detail later in this unit. Each succeeding version of the SCSI standard is backward compatible with all
preceding versions. In theory, you could put a SCSI-1 hard disk on a SCSI-3 host adapter. However, such
drastic implementations are not recommended because of complicated cabling adapters, termination
incompatibilities, and performance limitations.

105
SCSI-1
SCSI-1 started off as just “SCSI” but was later renamed SCSI-1 to avoid confusion with successive SCSI
standards. The following major features characterize the SCSI-1 standard:
• 8-bit parallel bus, meaning that multiple wires on the cable can transmit data at the same time.
This is what allows SCSI to use simultaneous I/O.
• 50-pin Centronics-style external connector and low-density pin header internal connector
• Single-ended (SE) transmission
• Passive termination (the simplest type of termination, but also the least reliable
• Optional bus parity checking
• 5 MHz operation
• Transfer rate of 4 MBps (asynchronous) or 5 MBps (synchronous)

SCSI-1 is now considered obsolete and has been retired by ANSI and replaced with SCSI- 2.

SCSI-2
SCSI-2 is essentially SCSI-1, plus the following optional features:
• Fast SCSI operating at 10 MHz instead of 5 MHz
• Wide SCSI utilizing 16-bit transfer instead of 8-bit (this is “narrow”)
• 50-pin high-density connectors
• Active termination (see later section on SCSI termination)
• High Voltage Differential (HVD) is used to extend bus length. HVD is now considered obsolete in
favor of Low Voltage Differential (LVD). (HVD and LVD are covered later in the unit.)
• Command queuing allows the host adapter to send as many as 256 commands to the drive. The
drive stores and sorts the commands for optimum efficiency and performance internally before
responding to the host adapter. Multitasking operating systems such as OS/2, Windows NT, and
Windows 2000 can take advantage of command queuing.
• Transfer rate of 10 MBps at 16 bits and 5 MHz (Wide SCSI), 10 MBps at 8 bits and 10 MHz (Fast
SCSI), or 20 MBps at both 10 MHz and 16 bits (Fast and Wide)

SCSI-3
SCSI-3 is the most confusing of the SCSI standards, because manufacturers have used misleading
marketing language. In addition, the SCSI-3 standard is not a complete standard of its own. Instead, it is
more a collection of documents covering new commands, electrical interfaces, and protocols. A
manufacturer could comply with only one of the major SCSI-3 additions and still label the product “SCSI3.”
Nonetheless, subdividing SCSI-3 into several smaller standards helps SCSI-3 implementations to develop
more quickly than waiting for the entire standard to be published and approved.

SPI SCSI-3 parallel interface

SPI is the SCSI-3 Parallel Interface, better known by the marketing terms Ultra SCSI or Wide Ultra SCSI.
Along with this standard is the separate SCSI Interlock Protocol (SIP) defining the parallel command set?
This standard was published as three documents offering the following features:
• 10 MHz bus speed, which is really no faster than SCSI-2. This is an example of how a manufacturer
could use the 10 MHz bus speed (Fast SCSI) and call it SCSI-3 even though it offered no
performance advantage.
• Fast-20 offers transfer rates up to 40 MBps using 20 MHz signaling.
• 68-pin P-cable and connectors for Wide SCSI

106
SPI-2
Also known as Ultra2 SCSI and Wide Ultra2 SCSI, SPI-2 is characterized as follows:
• Single Connector Attachment (SCA-2) connectors. A successor to the problematic original SCA
connector, this is the connection type that is mostly used in a chassis that contains several
hotswappable SCSI drives.
• Fast-40 40 MBps transfer rate on a narrow (8-bit) channel or 80 MBps on a wide (16-bit) channel.
LVD is required for these data rates.
• LVD signaling to replace the previous SE signaling is required to achieve the faster throughput of
Ultra2 or Ultra2/Wide speeds (40/80 MBps). You can use an SE device on the LVD SCSI chain, but
doing so switches the chain to SE mode, which throttles the performance of the chain to SE-level
performance at a maximum of 40 MBps and shortens the total cable length to as little as 5 feet
(1.5 meters) in Fast-20 mode.
• 68-pin Very High Density Connector (VHDC) makes the connectors and ribbon cables smaller. This
is important in SCSI because the older connectors and ribbon cables could be quite large and
cumbersome.

SPI-3
SPI-3 was still in the draft stage at the time of this writing, although SCSI manufacturers have already
implemented many of its features, including:
• Improving the earlier parity check, the cyclical redundancy check (CRC) is a calculation used by the
sending device based on the data in the packet. The data arrives at the destination target and
another calculation is performed using the same “formula.” If the calculation in the packet
matches the calculation performed by the destination device, the data is complete and considered
error free.
• Domain validation improves the robustness of data transfer. In the past, the host adapter would
send an inquiry to the SCSI device to determine its supported transfer rate. If the interconnection
between host and device did not support the full transfer rate, then the device became
inaccessible. With domain validation, the determined transfer rate is tested, and if errors occur,
the rate is incrementally reduced and again tested until no errors occur.
• Double transition (DT) clocking transmits data on both the rising and falling edges of the clock. On
a 16-bit, 40 MHz bus, this yields a transfer rate of 160 MBps.
• Packetization reduces the overall communication method to transfer data. Previously, data was
transferred over the SCSI bus using a series of phases to set up and transfer data. Packetization
streamlines this process by combining the process into a packet, reducing overhead.
• Quick Arbitration and Selection (QAS) eliminates the previously required arbitration method in
which devices contend for control of the bus. Much of the prioritization is based on the device’s
priority level based on its SCSI ID, a unique number for each SCSI device. Because no data transfer
can occur during the time that arbitration takes place, it adds overhead to the bus. QAS reduces
overhead by reducing the number of times that arbitration must occur and by allowing a device
waiting for bus access to do so more quickly.

SPI-3 is also known as Ultra3 SCSI.

107
Ultra160 and Ultra160+
The five features listed for SPI-3 sometimes because a problem in the way a SCSI product is marketed. A
manufacturer might include only one of the five items and call its product SCSI-3. Most manufacturers
prefer to have more stringent requirements, so for the Ultra160 and Ultra160+ standards, the product
must include:
• DT clocking at 160 MBps
• Domain validation
• CRC

QAS and packetization are optional for Ultra160, but required for Ultra160+.

Ultra320
Ultra 320 or SPI-4 doubles the transfer rate to a burst rate of 320 mb/s. This data rate is accomplished by
doubling the bus speed from 40 MHz to 80 MHz and using DT clocking. Like previous versions, it uses LVD,
packetization, and QAS.

SAS
The next step in SCSI evolution is Serial Attached SCSI, or SAS. Like PCI-Express and SATA, SAS abandons
shared parallel connections for serial, point-to-point technology. More like an Ethernet connection than
previous versions of SCSI, SAS requires just two pairs of wires to connect to each drive – one pair to
transmit and one to receive – providing full duplex communication at a steady rate of 300 MBps, with
higher speeds on the way. Fewer wires also mean much smaller cables and connectors compared to SCSI.
The cables and connectors are the same for SAS and SATA, and a machine with SAS technology can support
SATA drives also, though the reverse is not true.

iSCSI
Internet SCSI, or iSCSI, is a protocol that enables a computer to communicate directly with remote SCSI
storage over IP networks. SCSI commands from the computer are encapsulated in IP packets, given a
packet header, and sent out over Ethernet. At the other end, the commands are decoded and sent to the
SCSI controller. The process is repeated in the other direction as the data is sent back to the requesting
machine.

SCSI configuration
Configuring SCSI devices requires knowledge of the various types of SCSI, cables, capabilities, and
terminators. One of the most common sources of frustration and troubleshooting is an improperly
configured SCSI bus. When configuring SCSI from scratch, first consider the budget available to you. SCSI
can quickly become cost-prohibitive, especially with several high-performance hard drives. Closely
associated with the cost is the level of performance and amount of storage you want. Of course, the more
you want, the more you pay.

SCSI cables and connectors

Although you won’t see every possible combination of cables and connectors listed here, you should be
aware of those that are most commonly used. A narrow bus width (8 bits) is physically and logically smaller
than the wider 16-bit bus. The 8-bit, 50- conductor cable is also known as “A” cable and the 16bit, 68-
conductor cable as “P” cable. (Other cables have been proposed, but have not come to fruition because
the standards never caught on with manufacturers.) The “A” cable is a low-density cable with a Centronics-
style connector (see Exhibit 4-7) and 50 conductors. The “A” cable is used only in 8-bit (narrow)

108
implementations. If you have both narrow and wide cables, you can continue to use both standards with
special adapters such as the one in Exhibit 5-8, which reduces a 68-conductor cable to a 50-pin connector.

Exhibit 4-7: 50-pin, low-density “A” cable connector

Exhibit 4-8: 50-pin connector on a 68-conductor cable

Thankfully, in the SCSI-2 standard, a high-density 50-pin SCSI connector was introduced to save space,
because large SCSI connectors such as the Centronics type are cumbersome and can quickly take up space
in a cabinet. Also, instead of using the wire latches on either side of the connector to attach and remove
the connector, you use squeeze-to-release clip locks.

On the SCA-2 connector, the chassis has the female connector and the drive has the male connector.
(Recall that this is the connector that allows hot-pluggable hard drives.)

On the outside edges of both connectors are advanced grounding contacts that allow you to pull out or
plug in a SCSI drive without negative electrical consequence (see Exhibit 4-9).

Because you can’t see the actual connection take place inside the chassis, this is known as blind connector
mating. The connector provides both signal and power, and is an 80-pin connector offering only wide (not
narrow) SCSI.

Exhibit 4-9: SCA-2 SCSI attachment

Exhibit 4-9

109
With SCSI-3, you most often find the wide SCSI cable using a 68-pin connector both internally (inside the
case) and externally (outside the case and connected to the server’s SCSI host adapter).

Some SCSI connectors, though rare, are 25-pin connectors created for economy of both design and dollars.
This connector is streamlined compared to the 50-pin cable, and accomplishes this design mostly by
removing grounding—an inadvisable practice. Recall that grounding is a contributor to signal integrity.
Because the signal quality is not as good as standard 50-pin or greater SCSI cable, and because the
connector is visually identical to the commonly used DB25 parallel cable connector, this style of
cable/connector is not recommended.

Connectors are designed for either internal or external use. External cables are usually round with
thumbscrews or clip-lock connections on the connector. The round cable is highly shielded and well
engineered against signal degradation and interference. As a result, it is also relatively expensive.
Internally, you will see ribbon cable in any of several types.

Signaling
In order to obtain the correct accessories for your SCSI chain, you must first know what kind of signaling
is in use on the bus. Signaling generally describes transmission of data using electrical impulses or
variations. These electrical transmissions represent data that the sender originates and the receiver
translates based upon a mutually agreed-upon method. There are three types of signaling on the SCSI bus,
detailed in the following sections. You must be careful about mixing devices intended for one type of
signaling with devices intended for another type of signaling, because one or more devices might stop
functioning or become damaged. Also, you must know the signaling before you can properly terminate a
SCSI bus.

Disconnect-reconnect refers to the SCSI ability to be multi-tasking and multi-threaded. Either the initiating
or the target device can disconnect from the SCSI bus if it experiences a delay in processing a request,
then reconnect when it is ready to proceed. This keeps the bus free for other devices in the event of one
device’s delay.

SE
Single-ended (SE) signaling, the original signaling method used on the SCSI-1 bus, uses a common signaling
method in which a positive voltage represents a one and a zero voltage (ground) represents a zero,
resulting in binary communication. SE signaling is available for any SCSI-1 or SCSI-2 implementation as well
as SCSI-3 SPI-1. With most electronic signaling methods, you have a built-in opposition: the faster the
transmission speed, the shorter the maximum cable length. Of the three signal types, SE is most
susceptible to this limitation. On a Fast-20 bus, for example, an SE cable can only be 5 feet (1.5 meters)
long.

HVD
High Voltage Differential (HVD) signaling is also available for any SCSI implementation up to SCSI-3 SPI-1.
If you can use SE for the same SCSI implementations, then why use HVD? HVD, as the name implies, uses
comparatively more electrical power than the other two types of signaling, resulting in greater allowable
bus lengths. Whereas the SE cable is limited to 5 feet (1.5 meters) over Fast- 20, an HVD cable can reach
82 feet (25 meters). The “differential” in HVD represents a kind of signaling in which the signal is comprised
of the difference in a pair. A one is represented when one wire in the pair transmits a positive voltage,
and the other wire transmits zero voltage. The receiving device detects the “difference” between the
voltages, and translates it as a one. To transmit a zero, both wires carry a zero voltage.
110
This signaling method is much less susceptible to interference, signal degradation, signal bounce, and
crosstalk, allowing HVD to transmit over such a great distance.

Do not mix HVD and SE devices on the same bus. Because of the significantly higher voltage of HVD, SE
devices could get smoked—literally.
It is rare to find HVD in PC servers. It is more often found in older minicomputers and never gained the
popularity of SE, primarily because of a higher cost factor.

LVD
Low Voltage Differential (LVD) signaling is the signaling method you are most likely to find today, and it is
similar to HVD except, as the name implies, it uses a lower voltage.

Advantages of LVD are that you can use both LVD and SE devices on the same bus without electrical
hazard, and it allows a longer maximum cable length—up to 39 feet (12 meters). Many SCSI devices are
multimode devices; that is, they can operate as either LVD or SE depending on the signaling method of
the other devices on the chain. Multimode devices can be indicated in several ways, usually depending on
the preference of the company marketing the product. However, it is usually abbreviated as LVD/MSE
(where M is multimode) or LVD/SE.

Note the following about using LVD:

• A single SE device among LVD devices on the same chain will drop all devices to SE compatibility.
This could have a significant impact if the cable is over 5 feet (1.5 meters) when you add the SE
device, because the cable would now be too long.
• Some multimode devices have a jumper setting that forces SE operation. If you want to use LVD,
be sure that no devices have enabled this setting.
• Use either LVD or multimode LVD/SE terminators.
• You cannot operate both HVD and LVD on the same bus for reasons of electrical incompatibility,
which could damage LVD devices.
• LVD is the only available signaling method for Ultra3, Ultra160, Ultra160+, Ultra320, and probably
any other standards that emerge in the near future.

Because the connectors are identical regardless of the signaling method, look for a special symbol on the
connector. This will help ensure that you do not mix signaling standards, especially by adding HVD among
SE or LVD devices. The signaling symbols appear in Exhibit 4-10.

Single-ended SCSI devices

111
 Low voltage differential SCSI

Low voltage differential/single-ended multimode SCSI

High voltage differential SCSI

Exhibit 4-10: SCSI signaling symbols

SCSI termination
Termination (using terminators at the ends of a SCSI chain) is critical to assure error free operation on the
SCSI chain. A terminator absorbs the transmission signal to avoid signal bounce, making it appear to the
devices that the cable is of infinite length.

Terminators also regulate the electrical load, and are therefore critical in establishing a reliable
communications medium. Proper termination requires a terminator at both ends of the SCSI cable. Some
devices (and most high-performance host adapters) either automatically terminate or have a setting that
allows you to specify termination.

Otherwise, obtain a terminator to place over the last connector on the chain. If the last position is in use
by a device that does not terminate itself, you can place a terminator over the connection, which allows
signal transfer to and from the device while also providing the necessary termination. This is known as
pass-through termination (see Exhibit 4-11).

112
 Exhibit 4-11: Pass-through termination connects the chain on the side of the terminator and the device on the other side

The terminators must also have terminator power supplied, or they cannot properly terminate. If the host
adapter does not supply power to the terminator, you can configure a setting on one of the SCSI devices
(usually via a jumper) to supply power to a terminator. (Some administrators configure all SCSI
devices to supply power to terminators; that way it never slips through the cracks.) The
following list describes the various types of SCSI terminators:
• Passive termination—this is the simplest type of termination, but is also the least reliable. Passive
terminators use resistors to terminate the SCSI chain, similar to terminators on coaxial Ethernet
networks. Passive terminators usually work best on short, SE SCSI-1 buses. It is unlikely you will
find many passive terminators in servers.
• Active termination—A requirement for faster, single-ended SCSI, active termination adds voltage
regulators to provide a more reliable and consistent termination. Another type of active
termination is active negation termination, which uses a more complex circuit to stabilize the
voltage supply level, further eliminating electrical noise from the signal. Negation terminators are
usually only a little more expensive than plain active terminators ($5–10).
• Forced perfect termination (FPT)—in this technology, the termination is forced to a more exact
voltage by means of diode clamps added to the terminator circuitry. This advanced form of active
termination is very clean, and is the best termination available for an SE bus.
• HVD termination—a high voltage bus requires a high voltage terminator.
• LVD termination—A low voltage terminator for the LVD bus. Many LVD terminators are LVD/SE
terminators to accommodate buses with multimode devices. When operating in SE mode, the
terminator functions as an active terminator.

Be careful if your server environment mixes narrow SCSI technologies with newer, wider SCSI
technologies. First, to place both narrow and wide devices on the bus, you will have to provide an adapter
to convert from one width to the other. Converting wide devices to a narrow bus wastes 8 bits of
transmission and severely limits the potential of the wide device. On the other hand, converting from 8
bits to 16 bits does nothing to improve the performance of attached 8-bit devices because 8 bits is the
highest level of operation anyway. Along with the conversion, you must also consider how termination
will affect the two widths. Do you terminate for the 8-bit devices? If so, what about the 16-bit devices?
Terminating only for the 8-bits would leave the 16-bit devices with 8 un-terminated bits. These dangling
bits are called the high byte (also known as high 9), and when you have a mix of wide and narrow devices,
you can obtain special multimode terminators.

Multimode terminators flash colors indicating which signaling method is in use. For example, if it’s running
SE, it might blink yellow, and if it’s running LVD, it might blink green. This can be useful in troubleshooting
the SCSI chain if you wonder why it seems to be running slow or unreliably over 3meter lengths—it could
be running SE, (just check the light to verify) when you want LVD so that you can benefit from faster speed
and longer cable.
113
Drive configuration (SCSI ID and LUN)
Knowing the various SCSI technologies, cables, connectors, and termination is fine, but of course you have
to also plug it all in. The problem is that with SCSI, mis-configuration can cause many hours of baffling
troubleshooting. This section shows you how to correctly configure the devices on the SCSI bus.

Topology
Topology, the physical and/or logical layout of equipment, is simple for SCSI: Use a bus topology because
that’s all that’s available (logically similar to a network bus topology). Daisy chains your SCSI devices so
that they appear one after the other along the cable. In complicated configurations, you might accidentally
connect the bus to it somewhere along the line, creating a loop. Don’t do that or the bus won’t work.
When configuring a chain from scratch, one end usually connects to the terminated host adapter (unless
it’s between an internal and external chain) and the other end is terminated as well. The order of the
drives (and other SCSI devices, if any) doesn’t matter in terms of performance or priority. That’s one of
the advantages of SCSI—a slower device on the bus doesn’t bottleneck all the other devices. In a simple
configuration using two SCSI hard disks and a SCSI scanner (see Exhibit 4-12), the layout is a bus and both
ends are terminated.

Exhibit 4-12: A simple SCSI bus topology with terminators at both ends

Realize that although termination takes place at the end of the SCSI bus, the SCSI host adapter is not
always at one end of the chain. Because most host adapters have connections for both internal and
external devices, a host adapter so connected would appear in the middle of the SCSI chain. In this case,
you will have to be sure not to terminate the host adapter (see Exhibit 4-13).

You are likely to find a SCSI host adapter with multiple channels in many implementations. The primary
benefits to multiple channels are twofold: First, having two separate channels allows for twice as many
devices. Second, you can configure the two channels to support different signaling technologies. For
example, if you have a few SE devices and a few LVD devices, you don’t want to mix them or else you’ll
suffer the performance and length limitations of SE for all the devices. Instead, make one channel SE and
the other LVD. That way, the LVD devices can operate at full capacity.

114
 Exhibit 4-13: Do not terminate a host adapter appearing in the middle of the SCSI chain

SCSI ID Assignment
Each device on the SCSI chain must have a unique SCSI ID number. You configure the device SCSI ID using
jumpers, a wheel, or a button on the device, and the adapter is usually preset at ID 7, though you can
change it (possibly with a jumper but usually through software utility). Although you can change the host
adapter’s SCSI ID, it is highly inadvisable. ID 7 is the highest priority of all the SCSI numbers. The range of
available IDs depends on whether you are running narrow (8 IDs) or wide (16 IDs).

SCSI ID assignment is important not only to uniquely identify each device, but also to establish which
devices have priority when arbitrating over the bus. Arbitration occurs when it must be determined which
of two or more devices have control over the bus. Usually, you want to assign a higher priority to slower
devices such as scanners or tape drives to make sure that faster devices do not dominate the bus.
The priority of SCSI IDs range from the highest to lowest as follows: 7, 6, 5, 4, 3, 2, 1, 0. On a wide bus,
the range is 7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8. When you install an internal hard disk— simply
a matter of screwing it into an available drive bay and plugging in the power and SCSI cables— you also
need to configure a SCSI ID using jumper settings.

External SCA devices, however, usually configure their own SCSI IDs automatically as soon as you swap
the drive into its bay. This is needed to reduce the configuration time, as time is usually of the essence in
swapping these drives.

Along with the unique SCSI ID number is the logical unit number (LUN). The LUN is a subunit of the device,
and is used to identify items within the device. For example, a multidisc CD-ROM changer probably assigns
a unique LUN to each disk. If the CD-ROM drive is SCSI ID 5, the first LUN is probably LUN 0, the second is
LUN 1, and so forth. Some host adapters do not support LUNs—the absence of LUN support makes the
bus scan process during startup faster; so if you need LUN support, make sure the SCSI host adapter
supports it. Administrators are not normally concerned with changing LUNs because the device
manufacturer defines them.

SCSI pros and cons

SCSI has many pros and cons, but as mentioned earlier, the pros far outweigh the cons in most enterprises,
and particularly for servers.

Pros
The benefits of SCSI are as follows:

115
 • Performance—aggregating the performance of multiple disks as in a RAID array (covered later in
this unit) generates an appreciable performance gain. The more disks you add, the better the
performance (up to the available throughput of the SCSI channel).
• Expandability—several meters of cable length is enough for most SCSI implementations. Most use
LVD, which allows for 12 meters. Also, compared to ATA standards that limit you to two disks, the
expandability of using up to 15 devices on a SCSI chain is impressive.
• Redundancy—Several RAID implementations including RAID-1 and RAID-5 protect your data even
if a hard disk fails.

Cons
The drawbacks of SCSI are as follows:
• Difficult to configure and troubleshoot—SCSI has so many varying standards, and there are so
many opportunities for incompatibility and wrong termination, that it might take a while to get
SCSI rolling, especially if you are trying to piece together a SCSI implementation out of existing
equipment. If you build from scratch using consistent standards, it’s much easier.
• Expensive—Smaller organizations or those that have a tight budget might not be able to afford
SCSI.
• Performance—the performance improvement is not that noticeable if you’re only using a single
SCSI disk. SCSI shines when using multiple disks for performance and fault tolerance. Otherwise,
you’re better off saving the money and using ATA-4 or better.

Server administrators usually prefer SCSI because of its advantages in an enterprise environment.

4.4 FIBER CHANNEL AND RAID

Fiber Channel (FC) is a storage technology that can use gigabit Ethernet networks, but is primarily intended
for fiber optic cable as the name implies. (Fiber Channel technology has roots in Europe, hence the spelling
of “Fiber” instead of “Fiber.”) FC is a form of storage categorized under the storage area network (SAN)
umbrella. SAN is a general term that refers to any network-based storage solution that is not server-based.

SAN is becoming huge in the server world. Comprehensive coverage of SAN is beyond the scope of this
text because it encompasses more than just server issues.

First, let’s separate out from the discussion another growing storage solution that is sometimes confused
with FC—network attached storage (NAS). NAS is one or more storage devices attached to a network,
most commonly Ethernet. A NAS device is easy to configure and use, because you just attach it to the
rack, plug it into the network, flip its power switch, and it’s ready to use. NAS devices are as accessible as
any other device on the network. For smaller organizations, you can use a freestanding NAS device, or for
the enterprise, rack models are available. Regardless, when you attach and power up the device, it
receives an IP address and identifies the type of network in use.

If you want to configure a static IP address, configure security settings or access privileges, or configure
RAID, you can launch a management utility (usually browser based so you can launch it from anywhere in

116
the enterprise) and make changes. NAS devices are very inexpensive compared to FC. NAS devices as well
as SAN storage have a significant benefit over traditional file servers in that you don’t have a file server.

This removes an expensive piece of hardware and greatly reduces administration.

In a SCSI FC implementation, a special SCSI host adapter designed for FC connects to the FC bus, which
can be up to 10 kilometers (more than 6 miles) in length with single mode fiber optic cable (multi-mode
cable allows for 500 meters between devices).

That’s the benefit of using fiber instead of copper. Network adapters can connect to fiber with the use of
a Gigabit Interface Converter, or GBIC. Small Form-factor pluggable GBIC’s (SPF GBIC) are used in network
switches.

FC is run in its own storage “network” of either Fibre Channel Arbitrated Loop (FC-AL) or switched fabric.
In FC-AL, you can connect 126 storage devices to a special fiber hub in a physical star, logical loop topology.
However, all devices on the FC-AL share the available bandwidth, which is OK in many implementations if
it meets the needs of the enterprise (see Exhibit 4-14).

Exhibit 4-14: An FC-AL configuration connects to a fiber hub and shares available bandwidth

In high-level implementations, you can use a switched fabric to connect to up to 15.5 million storage
devices, and the full bandwidth of the channel is available to each device. As you might guess, this
implementation is extremely expensive. The term switched fabric is somewhat vague in terms of its
physical configuration. Simply put, it means that servers storing information on an FC device can reach
the device using any number of physical paths. This is analogous to a phone call to a friend in another
state. Call her today, and the phone company will route your call using the best path it deems practical.
Call her tomorrow, and the phone company might or might not use the same path. However, it doesn’t
matter to you or to her, so long as the call connects. Some people have prematurely dismissed FC as too
slow at its rated throughput of 100 MBps (200 MBps full duplex), especially compared to Ultra160, which
can reach up to 160 MBps. In reality, FC is much faster because SCSI cannot sustain speeds of 160 MBps;
that’s just its burst speed when feeding data from cache. Instead, it will usually dish data out at a sustained
rate of around 50 MBps. FC, on the other hand, can sustain data transfer at about 97.5 MBps.

The storage media can be whatever file format suits you, and you can use RAID if you like.
117
Developments in fibre channel now allow for transfer speeds of 10 Gbps.

Typically administrators do not purchase the FC equipment and configure it themselves.

Instead, you contract with a SAN storage company specializing in FC to supply the equipment and connect
it all together. Maintenance is not usually an issue either, as the storage company also provides
maintenance. Larger enterprises often have full-time employees from the storage company on site at all
times to maintain and administer the storage and tapes.

The expense of FC is very high, and thus is usually seen in enterprises that need to store terabytes of data,
such as in a large data center. The storage devices for a single terabyte of data alone can cost up to
$750,000 before installation and costs of other associated equipment, with many implementations
totaling millions of dollars.

FCIP
Like iSCSI, Fibre Channel over IP (FCIP or FC/IP) enables fibre channel storage to operate of IP networks,
allowing for geographically distributed storage facilities.
RAID

RAID stands for either Redundant Array of Inexpensive Disks or Redundant Array of Independent Disks. It
doesn’t really matter—it’s just that the history of RAID has changed over the years and the acronym has
changed with it. What matters is how you use RAID in the enterprise, and that you understand the
characteristics of several levels of RAID, particularly RAID-0, RAID-1, RAID-5, and RAID-0+1. The main
purpose of RAID is to use multiple disks to improve performance, provide redundancy, or both. You
configure a RAID array via host adapter hardware or software, though in most cases hardware RAID is
preferred for reasons we will discuss at the end of this section. To the operating system and applications,
the drives are logically a single drive.

With a controller that supports it, you can obtain a significant performance benefit using a RAID cache.
Regardless of the version of RAID you use, the host adapter can have a certain amount of memory, usually
a minimum of 32 MB, though better adapters have at least 128 MB. The memory is often supported by a
battery backup for data integrity, similar to CMOS. The RAID cache fills with data sequentially beyond the
actual requested data in anticipation that the next data will soon be requested. If the data is indeed
required, the RAID cache serves data more quickly than if data must be retrieved directly from disk.

RAID-0
RAID-0, also known as disk striping, lays down data across two or more physical drives. RAID-0 provides
no redundancy—if one of the drives fails, all the data is lost, so you must provide another solution for
redundancy such as tape backup or RAID-0+1. RAID-0 yields a performance advantage because you can
aggregate the performance of multiple drives. The SCSI bus can deliver data in a matter of microseconds,
but the data transfer of a single disk takes milliseconds, hence a bottleneck at even the fastest drives.
RAID-0 helps to mitigate this bottleneck because multiple disks can simultaneously deliver data at once.
For example, a RAID-0 array consisting of four drives provides roughly four times the performance of a
single drive serving the same data because all four drives on a SCSI chain can operate at once. In Exhibit
4-15, the letters on each of the drives represent a portion of data, illustrating how the data appears on
the drives. In any of the RAID levels, you can further improve performance by dedicating a single drive to
each host adapter. (Using two host adapters and one drive on each is known as duplexing.) It is possible
118
to use RAID-0 in an IDE implementation, but because IDE can operate only one drive at a time, there is no
performance advantage.

Exhibit 4-15: This RAID-0 array uses four drives to quadruple hard disk performance

Part of the POST process involves searching for configurable BIOS from other hardware, particularly SCSI
host adapters. To configure a RAID-0 drive once the proper SCSI connections have been made, enter the
SCSI adapters BIOS. Typically, a message appears instructing you to press a keyboard combination to
access the BIOS. Then, use the menu system to specify the type of RAID you want to use and save changes.

RAID-1
RAID-1, also known as disk mirroring, requires at least two disks to provide redundancy. It can also provide
improved performance, but only if using duplexing. In a RAID-1 array, the controller writes the same exact
data to two disks at the same time. You can configure multiple adapters, each having two channels with
a disk on each channel and capable of simultaneous I/O, to further increase performance (see Exhibit 4-
16).

Exhibit 4-16: RAID-1 using multiple controllers provides both mirroring and high performance

RAID-1 has a higher cost factor or “overhead.” If you have two mirrored 40 GB drives, the total raw storage
available is 80 GB, but you can only use 40 GB because 40 GB must be available for the mirror. The
overhead is 50% because you only use half of the actual disk space. Also, if the drives are not the same
size, the mirror is the size of the smaller drive. You can configure RAID-1 through software (using Windows
NT/2000, for example), but it requires processor utilization; so in performance terms you’re better off
using hardware RAID through the controller.

In the event of a failed drive in a RAID-1 array, simply pull out the failed drive and insert a new one.
Depending on the method used to configure RAID, you must manually regenerate data from the remaining
drive to the new drive through software or BIOS, or the mirror regenerates automatically. With automatic

119
regeneration, the user experience is not disrupted, although performance may suffer temporarily until all
data is regenerated to the new drive.

RAID-0+1
As the name implies, RAID-0+1 offers the best of both worlds: the performance of RAID-0 and the
redundancy of RAID-1. In this implementation, two channels and at least four drives are required. Data is
striped across two or more disks in the first channel (RAID-0), and the data from the first channel is
mirrored to disks in the second channel (RAID-1) in the same striped layout as shown in Exhibit 4-17. This
implementation has a 50% overhead.

Exhibit 4-17: A RAID-0+1 array

Contrary to some sources, this is not the same as RAID-10. Although RAID-10 also uses both mirroring and
striping, one channel mirrors the data and the other channel stripes the same data. In RAID-10, overhead
is higher, although two disks could fail and you would still have enough fault tolerance to rebuild the data.

RAID-3
RAID-3 uses striping like RAID-1 but also adds redundancy by use of parity. In the case of RAID, parity is an
encoding scheme that represents data appearing on other drives. RAID-3 saves the all the parity
information to a single parity drive. The more commonly used RAID-5 spreads the parity and data across
all the drives in the set.

RAID-5
RAID-5, sometimes called striping with parity, requires at least three hard disks to implement. The host
adapter writes data to drives 1 and 2 and on disk 3 writes parity data. At the next write, data writes to
drives 2 and 3, and on disk 1 writes parity data, and so on in round-robin fashion (see Exhibit 5-18). If the
first disk fails, replace it. Then, the system regenerates the data that was supposed to be on the first drive
using the parity on disk 2 and disk 3. SCSI is the best choice for this RAID implementation because it offers
simultaneous I/O; that is, data reads occur from multiple drives at the same time.

RAID-5 has a lighter overhead in terms of disk space than RAID-1. Overhead is 1/X where X is the number
of disks in the array. For example, a five-disk RAID-5 array would have an overhead of 1/5, or 20%. The
overhead space is utilized by the parity data.

RAID-5 arrays can take quite a while to rebuild data to a new replacement drive. Also, parity calculation
requires CPU cycles, and can significantly affect server performance, especially while rebuilding data to a
replacement drive.

120
Exhibit 4-18: RAID-5 writes parity bits across members of the array, and uses the bits to reconstruct data if a drive must be
replaced

RAID-1 and RAID-5 reconstruct data with the least amount of disruption if the hard disks are hotswappable
and data is reconstructed on the fly as opposed to rebooting and accessing the BIOS to reconstruct the
data. (Although manufacturers such as Promise Technology offer EIDE host adapters providing RAID-0, 1,
5, and 0+1, performance benefits are best with SCSI because EIDE cannot perform simultaneous I/O.)
Many RAID systems support hot plugging or hot swapping of drives, meaning a failed drive can be pulled
from its slot and replaced, and the new drive rebuilt, without ever turning off the machine or interrupting
network service.

Regardless of the RAID configuration you choose (if any), be sure to optimize the hard disk so that it
performs at its best. One of the most useful utilities for this purpose is defragmentation software, which
arranges the data on your hard disk in a sequential fashion so that reads require less back-and-forth action
to locate all the data.

Other RAID combinations

Like RAID 0+1 and RAID 10 described above, other RAID levels can be combined to achieve greater levels
of capacity, speed, or redundancy. For instance, in RAID 5+1, entire RAID-5 striped sets are mirrored, while
RAID 0+5 is a RAID 5 array made up of RAID-0 sets.

Software RAID vs. hardware RAID

Network operating systems such as Windows NT/2000 and Novell NetWare offer software RAID.

Software RAID Advantages

• Software RAID is built into the NOS, and is less expensive than hardware RAID.
• Generally an easy-to-use graphical interface. You can create a RAID array in a matter of minutes,
perhaps without rebooting the system.

Disadvantages
• Less robust than hardware RAID. If the operating system becomes corrupt, the array is at risk.
• Limited configuration capabilities. You are usually limited in the types of RAID configurations you
can use. In Windows 2000, for example, you can create a RAID-0, RAID- 1, or RAID-5 array but not
other RAID configurations such as 10 or 0+1.
• Uses the CPU to perform parity calculations.

121
Hardware RAID Advantages
• More reliable and robust Faster performance because it is dedicated to RAID functions.
• Also, parity calculations are handled by hardware RAID (not the main CPU), which is faster than
software RAID.
• The array remains regardless of corruption to the operating system.

Disadvantages
• More expensive
• If you need to access the SCSI BIOS, a reboot is required.

All things considered, experienced administrators usually recommend hardware RAID as a better solution
because of its dependability and performance.

Zero Channels RAID

A zero channel RAID card, or ZCR card, makes use of a system board’s existing SCSI controllers. The ZCR
card doesn’t have all the chips necessary to support RAID on its own. This lack of an on-board SCSI chip
and connectors makes the card much cheaper and more compact.

Review Questions

1. Why can larger cluster sizes waste disk space?

2. In what ways does FAT32 improve over FAT?
3. What limits the file size in NTFS?
4. What file systems are used by UNIX and Linux?
5. Which operating systems support drives larger than 8.4 GB?
6. What is an advantage of IDE/ATA?
7. Which two ATA implementations can use 100 MBps?
8. What is a synonym for an IDE or SCSI controller?
9. What does a SCSI parallel bus allow?
10. Which version of SCSI introduced both Fast and Wide implementations?
11. A Wide SCSI implementation would use how many pins?
12. What does an SCA-2 connector allow you to do?
13. How does double transition (DT) clocking improve SCSI performance?
14. Why is a SAS cable smaller than a SCSI cable?
15. Why shouldn’t you mix HVD devices with LVD devices?
16. What is the best form of termination for an SE bus?
17. What is the SCSI ID of most host adapters?
18. What is one major drawback to SCSI?
19. What’s the difference between SAN and NAS?
20. What’s the maximum fibre channel bus length with single mode fiber-optic cable and with
multimode?
21. What’s FCIP?
22. Why use RAID?
23. What are the advantages of software RAID?
24. What are the advantages of hardware RAID?

122
 TOPIC 5
_____________________________________________________________

5. SERVER UPGRADES
_____________________________________________________________

LEARNING OUTCOMES
After studying this topic you should be able to:
• Prepare for a server upgrade, verify availability of system resources, and adequately test and pilot
the server upgrade.
• Upgrade the processor and memory.
• Upgrade the BIOS, power supply, UPS, and adapters.

5.1 PREPARING FOR A SERVER UPGRADE

As an administrator, you must execute every part of an upgrade with the utmost care and sensibility. First,
you probably have to obtain approval to perform the upgrade, which involves budgeting, planning, and
presenting your plan to others. Even if you have autonomous power to make decisions for server
upgrades, you still need to plan carefully, because if something goes wrong, you are the responsible party.
You must also time the upgrade so that it has a minimal impact on productivity and user experience. Be
sure to educate users as to what to expect regarding any upgrades that directly affect them. Finally, before
performing the physical upgrade, verify that all the server components are there and in working order.
With proper planning and precautions, you are now ready to perform an upgrade.

When to upgrade the server

On a simple level, an administrator might perform an upgrade in just minutes without notifying anyone.
For example, perhaps a server just needs more memory costing less than $200. In many organizations,
administrators have discretionary spending for smaller amounts, and it should only take a few minutes to
shut down, install the memory, and restart the server. However, upgrading a server can require a
significant investment in time, planning, and, of course, money—especially for larger upgrades that have
a significant impact on the operations of the IT department, users, or your business audience.

For example, if an organization primarily employs eight NetWare 3.12 servers and would like to upgrade
the servers to Windows 2000 Server, it is likely that the new system will need significantly more powerful
hardware, as the minimum system requirements for NetWare 3.12 are much lower than for Windows
2000. (Granted, this is a drastic upgrade, but it makes very clear the effect that an upgrade can have on a
network.) The organization will likely have to upgrade all major components, including the processors,
memory, hard disks, and possibly even the motherboards. A mass migration and upgrade like this would
probably be more cost-effective if the servers were replaced with new ones.

Moreover, this upgrade affects more than just how the IT department administers the servers—network
resources for users might be located differently. Logon scripts mapping to NetWare printers won’t work
anymore, and users that have manually created shortcuts to network resources will no longer be able to
access those resources. The impact of an upgrade this drastic would require significant planning, proper

123
timing, a smoothly executed upgrade process, adequate personnel, and plenty of communications with
the users notifying them of the upgrade and its impact on their day-to-day functions.
Many times, upgrading the server is a necessary step in response to poor server performance. To fully
justify expenditures in time and money, you should create a performance baseline so that you can define
an acceptable level of performance. If you see that the server begins to perform poorly on a regular basis
when compared to the baseline, consider upgrading components that might be a bottleneck. For example,
a heavily utilized database server will have higher demands placed on the processor than a simple file
server. Upgrading or adding another processor might help the system to perform better.

Before upgrading a server component, pause to verify identification of the actual bottleneck. For example,
a server that shows heavy hard disk utilization might not need a larger or faster hard disk. Instead, first
check memory utilization. Most NOSs utilize a swap file mechanism that substitutes hard disk space for
memory in a low memory situation. Therefore, heavy hard disk utilization could actually be a memory
issue that would be minimized with a memory upgrade. You might take that same database server and
consider upgrading it not only to an acceptable level of performance, but beyond current needs in a
proactive approach to extend the server investment. This might prevent the repeated expense of
upgrading again in a few months as demand on the server grows. For example, if the database server has
one 700 MHz Pentium Xeon processor now, you might consider installing two 1 GHz processors, if the
motherboard supports it. A proactive upgrade such as this is sometimes also in anticipation of financial
timing. For example, in some organizations, if the department does not use its entire budget in a given
time, then the next budget allocation might shrink because of a perception that the department does not
need as large a budget since it didn’t spend the entire budget last time.

Although this kind of budgeting model has obvious weaknesses in logic and wisdom, it is nevertheless a
reality in many organizations.

Other reasons for a proactive approach might include the anticipation of an upcoming merger with a new
parent company. Perhaps your current intranet is for general information only and does not have a high
hit rate. However, the new parent company requires you to post a great deal of company information on
the intranet, such as details on employee benefits, company announcements, Webbased collaboration
software, and so forth. In this case, you might consider several upgrades, including adding another NIC to
the server, to increase network throughput in anticipation of a higher hit rate.

Timing the upgrade

Generally, the IT professional tries to keep as low a profile as possible within his or her organization. Why?
Because when the administrator has a high profile, it’s usually because something is wrong with a server
or the network, and users tend to immediately think of administrators as having caused the problem.
When users don’t notice your presence in the organization, it usually means you are doing your job well.
An ill-timed server upgrade (or scheduled maintenance) can give the administrator a very high profile;
therefore, you should make sure that day or weeks before the upgrade, you monitor server utilization to
determine periods of peak usage and periods of lowest usage. Of course, you want to perform any
upgrades that make the server unavailable only during periods of low usage. If your organization is most
active during business hours, you can stay late or arrive early to perform your upgrades while nobody else
wants to access server resources. However, many organizations, especially larger ones, operate
continuously. In addition, if the organization is global, users on one side of the globe might access
resources while users on the other side of the globe are sleeping. This type of continuous access means
that some types of upgrades will definitely take place while some users are attempting to gain access to
resources. If the server must be taken offline, you can either temporarily transfer the role of server to
124
another server or you can notify users in advance so they don’t expect access during the time of the
upgrade. Some upgrades (such as adding a disk to an array) might not require any downtime, and users
will never notice the difference. On the other hand, upgrades such as adding processors or memory
require the server to be turned off, and you should schedule an appropriate time and notify users in
advance.

Notifying users
Notifying users of a server upgrade helps to reduce the administrator’s visibility and avoid unnecessary
calls to the IT department. Even if you try to upgrade the server during the lowest usage periods, someone
will still wonder why the server is unavailable, so the notification should help. Also, a public advance notice
such as an email broadcast shows that you made a reasonable effort to notify users. Broadcast company
e-mail is a common notification method, as are notices on the company Internet or intranet. Start
notification as far in advance as is practical, which will vary from one organization to the next. (Some well-
organized organizations have a written administrative policy for planned downtime notification.) An initial
notification far in advance is recommended; and then as the upgrade approaches, notify a few more times
with increasing frequency up until the actual upgrade event.

Notify the users when the server will be unavailable and for how long. If possible, also state how the
planned downtime benefits the user. This helps to psychologically cushion the inconvenience for users
when they understand it is ultimately for their benefit. For example, most users would be grateful if you
were to add more storage to the e-mail server so they could store more messages.

When a server goes offline unexpectedly and not as a result of upgrades or maintenance, you should
attempt to send out a message to all users to stem the certain flood of calls about the inaccessible server.

Try to maximize the upgrade process so that other tasks can be completed in the same approximate time
without increasing the impact on user access. For example, any time you want to work inside the server
case, you might as well get a vacuum or a can of compressed air and eliminate the dust. Since that’s
probably also a regularly scheduled maintenance item, doing it now saves the separate task of doing it
later.

Confirm that you have necessary upgrade components.

Avoid unproductive downtime during an upgrade by confirming that all the necessary components for a
successful upgrade are accounted for. For example, some hardware might not respond appropriately to
the operating system without the proper BIOS upgrade, so you need both the BIOS upgrade and the
hardware itself. When you perform an upgrade, be careful about the drivers that come with the hardware.
(A driver is a software interface that allows the hardware to function with the operating system.)

You do not know how long the device has been on the shelf, and the drivers might be outdated. Save
yourself the task of time-consuming troubleshooting after the upgrade that might occur due to an
outdated or incompatible driver. After checking to see if the device is compatible with the other hardware
and network operating system (NOS), download the most recent driver from the vendor’s Web site. While
you’re there, check the FAQ section to address and prepare for any issues you might encounter during the
upgrade.

Download and expand the drivers into a permanent network directory from which you install the drivers.
You might need to reinstall the drivers from time to time, and many NOSs default to installing drivers from
125
the original installation path. Also, the drivers are immediately available to other servers on which you
perform the same upgrade.

Make sure you dedicate each directory to only a specific vendor and a specific device, because driver files
might have the same name (especially from the same vendor) and you might accidentally overwrite files.

Similarly, check with the NOS vendor to see if there are known problems with the particular device and/or
driver you want to install. A Readme.txt file often accompanies drivers and updates. Although this file is
usually a statement of obvious information (“this driver upgrades your network card”), it might also
contain important information (“this update only applies to Windows 2000”). Many administrators ignore
this little file, but at the very least, you should scan it for any red flags or installation tips. The NOS Web
site often informs you of incompatible devices and provides solutions. Often, the solution is to avoid
specific conflicting hardware devices or to install a NOS upgrade, patch, or hot fix. If the NOS vendor’s
support team does not list any known problems, check with other sources, such as newsgroups focused
on the specific hardware and/or NOS vendors.

The level of precaution recommended in these pre-upgrade tasks might seem overly cautious. If you were
performing a simple upgrade on a home PC, it might be. However, because of the impact the server has
on an organization, every precaution is necessary.

Planning for a failure

Anybody who has administered computers for more than a few weeks knows that even the most
meticulous and professional upgrade attempt sometimes encounters unanticipated problems, conflicts,
or incompatibilities. In preparing for this possible contingency, administrators should prepare adequate
failsafe measures to quickly recover from the problem or continue service to users through redundancy
while troubleshooting the failed upgrade.

Always back up the server before performing any hardware or software upgrades. Do not depend upon
the normally scheduled backup rotation, because if a problem occurs, the backup can be slightly outdated.
For example, most backups take place in the middle of the night, but if you perform the upgrade after
everyone leaves work but before the backup, then one day’s working data is in jeopardy if a problem
occurs. Instead, take the server off the network so that new data cannot be written to it. Next, perform a
full backup of at least the data, and possibly also the operating system. Now you have a snapshot of the
system before the upgrade, and if necessary, a restore should replace the data intact.

Consider using imaging software such as Power Quest Drive Image or Norton Ghost.

This exactly duplicates the hard disk to a single image file for restore should a problem occur. You can
store the image on a network share or one or more CD-ROMs, and protect it with a password in case the
CD-ROM falls into unauthorized hands.

Alternatively, many network cards are now bootable, allowing you to access the network and restore the
image to the server from the network share. If you have to restore the image, it is much faster than
reinstalling the operating system, reinstalling all of the server applications, and then restoring the data
from a conventional backup.

In addition to creating a backup, other failsafe methods can also ensure server availability. For example,
recall that clustering is utilization of two or more servers hosting the same application. If one of the servers
126
is unavailable (as might occur during an upgrade), the remaining servers in the cluster continue to provide
service. With a mission-critical server or application, you probably already have clustering enabled. Absent
a cluster, administrators might have a hot spare, which is a specific component (usually a hard drive) or a
complete server that can immediately be available on the network and transparently perform the exact
same functions as the original.

Verifying system resources

You never seem to have enough PCI slots for the devices you want to install in a server.

While you might have a free ISA slot, you probably don’t care because fewer devices are ISA compatible.
However, PCI slots represent valuable slot real estate that quickly fills up. In a typical server, you probably
have five PCI slots. Account for two network cards for better throughput and availability, a SCSI card for
tape devices, and another for hard disks, and you’re almost out of expansion slots already. If so equipped,
you can add a mezzanine or riser board to the motherboard to expand the number of available slots.
Whatever the case, in larger environments the administrator probably does not know offhand exactly how
many expansion slots remain in each server. Before making plans to add a device, verify that sufficient
slots are available.

Even if a slot is available, you might encounter issues with available IRQs, DMAs, or I/O ports. These are
each limited resources that most devices require to communicate with the operating system and other
devices, and they are defined as follows:
• An interrupt request (IRQ) is a request that the device uses to “interrupt” the processor to ask for
processor resources. There are 16 IRQs, numbered 0–15.
• Several IRQs are pre-assigned. For example, the COM1 serial port usually has IRQ 4. PCI IRQ
steering can allow multiple devices to use the same IRQ if no more unique IRQs are available.
However, ISA devices cannot take advantage of this benefit.
• A Direct Memory Address (DMA) is a resource that ISA devices use to directly access memory
without first having to access the processor, both increasing device performance and reducing
processor load. There are eight DMA channels, numbered 0–7.
• An I/O port is a location in memory that the processor uses to communicate with a device.
• A memory address is a dedicated region in system memory that some devices reserve and that is
unavailable for use by any other device, application, or the operating system. This can help device
stability by ensuring that nothing else trespasses the memory, which causes system errors.

If you are out of IRQs and IRQ steering is not available, or if you are out of available MAs, then you must
remove (or disable) an existing device that requires those same resources, or you cannot upgrade the
server. I/O ports are usually plentiful, and if two devices request the same I/O port, you can usually
reassign one of them to an alternate port.

Depending upon the chassis and power supply, you might not have sufficient expansion space to add more
hard disks, tape drives, or removable storage such as Zip drives, CDROMs, or DVD devices, all of which
require a drive bay, either internal or external.

Drives installed in an internal drive bay are neither accessible nor visible when the case is attached.
Most commonly, you install hard disks internally. It would be impractical to install removable storage or
CD-ROM/DVD drives internally, so you want to use an external drive opening, which means that you can
access and see the drive. Besides drive bay availability, you need a power supply that can handle the

127
additional power requirements of the devices and has sufficient power connectors for each drive. You can
use a Y-cable split that converts a single power connector to two, but if you’re using several of these, you
might be overloading the power supply. Generally, the more powerful the power supply, the more power
connectors it includes.

A standard desktop or entry-level server tower case has space for perhaps four internal drives and two or
three external drives. A server has significantly more storage space.

For example, the Compaq ProLiant 8000 has internal drive cages for 21 hot-plugs hard disks. Regardless,
verify available drive bays as necessary.

If you plan to install a large quantity of internal hard disks, consider adding one or more additional cooling
fans to compensate for the additional heat.

Making an inventory
One of the most frustrating things about installing hardware is finding it. Even relatively small
organizations quickly accumulate quite a few loose components, chips, hard disks, and so forth. When a
server fails, you must know exactly where to find replacement parts, so you should carefully inventory
(and lock up) all parts that have any value.

Organize smaller loose parts in appropriately sized storage bins, trays, and cabinets.

Having an inventory also helps you to control and be aware of possible theft.

In addition, you should know what equipment is in each server for proper asset tracking, budgetary
projections, and warranty service. Especially in large environments, manual inventory of installed
hardware is an arduous and seemingly endless task, further complicated when there are multiple sites.
You might want to procure software that can automatically scan your entire network to inventory not only
installed hardware in your servers and clients, but also installed software. Some programs can also identify
network devices such as hubs, routers, switches, and so forth. One of the most popular programs is
Microsoft Systems Management Server, for which you can find detailed inventory instructions at
www.microsoft.com/technet/SMS/c0318341.asp or perform a search for the title “Administering
Inventory Collection.” Also check into the following other vendors:
• Hewlett Packard’s OpenView at www.openview.com
• IBM Tivoli at www.tivoli.com
• Computer Associates Unicenter TNG at www.cai.com

Note that all these products require hardware that is capable of responding to queries from the software.
You can still manually inventory hardware that does not automatically respond to the software, but this
is becoming less of an issue as more hardware is designed to be compatible with inventory software.

When receiving new equipment or equipment transferred from another office, always request that an
inventory list be included with the shipment. This helps to ensure that equipment arrives as promised and
that the server from the home office that has 512 MB RAM doesn’t suddenly appear in your office with
only 128 MB. Another reason for the inventory list is that it helps you in assembling the equipment. Many
servers and their associated equipment involve dozens of parts, including zip ties, cable management
systems, fans, screws, keys, books, warranty cards, power cords, and so forth. It is extremely frustrating

128
to unpack and assemble an entire server and rack, only to find that you are missing a vital component that
you would have known about had you compared the physical parts to an inventory list.

Test and pilot

The potential impact of some upgrades (both if they succeed and if they fail) might require an isolated
pilot program in which you thoroughly test the upgrade for reliability and performance prior to
deployment throughout the organization. A pilot program isolates a server upgrade in a portion of the
network that makes performance easier to determine and lessens negative impact should some part of a
major upgrade fail. Some organizations require a pilot program for nearly any hardware, operating
system, or software change. However, pilot programs are not usually intended for common upgrade items
such as installing a hard disk, network card, or memory. Most implementations might involve something
like an upgraded NOS, an entirely new NOS (migrating from UNIX to NetWare 5.x, for example), or a
change in hardware architecture (such as a change from UNIX-based Alpha architecture to Windows-
based Intel architecture). If the upgrade works well in its initial pilot, you can extend the pilot programs
to other segments of the network to see if the results are also successful under different circumstances.
Finally, when the upgrade is fully tested and has satisfactorily passed the pilot phase, you can deploy the
upgrade throughout the remainder of the organization as necessary. As an example for the current
context, which focuses mostly on hardware, consider a pilot program for upgrades of the BIOS,
motherboard, processor (especially if changing platforms from, say, AMD to Intel), and anything else you
think might significantly impact the network if it fails or requires isolated analysis.

Once you successfully implement the upgrade changes, make sure that the improvement is more than
just a perception. Start recording the performance of the server and/or network as it applies, and compare
it to the previous baseline and performance prior to the upgrade. You should be able to find an
improvement in the targeted upgrade area. For example, if you added another processor, you should see
overall processor utilization drop to a lower percentage (where lower percentages equal better
performance). After recording performance statistics, you should be able to change the baseline’s level of
acceptable performance. In most organizations, the baseline is a moving target that you periodically reset
as server demands increase. If responsible parties (you and management) determine that network or
server responsiveness no longer meets an acceptable range of performance, then a decision must be
made: Either reset the baseline at the new level of performance or modify network or server equipment
to return to the original (or better) level of baseline performance.

When confirming a successful upgrade, you can check for the obvious items such as proper functionality
of the hardware. Don’t forget, however, that most NOSs have logs that might also record errors in a
problematic upgrade.

Be sure that after performing the upgrade, you record it to an easily accessible source for troubleshooting
and asset-tracking purposes. Some organizations might have a log book next to the rack where changes
are hand-written; others might have a computer-based log, such as a database or a spreadsheet, saved to
an administrative network share.

For many administrators, the latter is preferable because the log is accessible from any administrator’s
desktop. For example, in troubleshooting a server in Denver, an administrator in Phoenix can open the
server log on a network share and see if any recent hardware upgrades might have caused a problem.

129
5.2 UPGRADING THE PROCESSOR AND MEMORY

When you’ve completed all the steps outlined here to prepare for an upgrade, you should be ready for
the hands-on work of upgrading a server or server components.

Before you start, though, be sure that you’ve taken precautions to prevent electrostatic discharge.

Avoiding electrostatic discharge

Before touching anything inside the server, it is critical to exercise precautions against electrostatic
discharge (ESD). You have probably experienced ESD at no significant harm to yourself many times,
particularly if you live in an area where winters are cold and the wind blows (Chicago, for example). Once
you are in from the cold and touch a door handle (or pet the cat), an ESD occurs. ESD occurs when two
objects with differing electrical potential come into contact with one another because the electrical
charges seek to equalize. While you are outside in the cold wind, the energy from the wind can build up
electrical potential in you. When you touch an object in the house with less electrical potential, static
electricity discharges from you to that object. Although people are hearty enough to sustain a static shock,
servers and their components are not. Before you touch a server for any reason, including component
installation or inspection, you must be certain that you present no ESD threat. Although it doesn’t seem
like much, if you can feel the static discharge, then you probably discharged around 3500 volts. If you also
hear the discharge, then you probably discharged around 5000 volts, and if you see the discharge in a
lighted room, then you probably discharged around 8000 volts. It is easy to build up this voltage— walking
across the floor can generate 15,000 volts, and removing bubble pack from a carton can generate as much
as 26,000 volts. It only takes 100–1000 volts to negatively impact server components, and at lower levels
you might not feel, hear, or see the discharge.

You might unknowingly damage a component with ESD in two ways. First, an upset failure affects only the
reliability and/or performance of a component. This is perhaps the worst of the two types of damage
because it is difficult to detect and consistently reproduce. (For example, it takes 200–3000 volts to
damage a server’s CMOS.) The second way ESD affects a component is a catastrophic failure, which
immediately damages the component so that it ceases to function properly.

In seeking to prevent static discharge, take the following precautions:

• Touch the chassis—touching the chassis grounds you and equalizes the voltage levels between you
and the server. This is not a sufficient precaution, however, because as time passes, voltage can
build up in you again, particularly if you are wearing leather soles on carpet. Touching the chassis
is only a temporary, initial precaution. If you cannot utilize a better method, such as using an ESD
protection kit, then continue to touch the chassis periodically.
• Unplug the power—many people assume that because the power cord is plugged into a grounded
outlet, the case is protected against ESD. While it is true that the plug does lead to earth ground
that is not what is important in preventing ESD. What you are seeking is equalization in electrical
potential. Plugged or unplugged, touching the chassis temporarily equalizes electrical potential.
Moreover, it is safer to unplug the server. By leaving the server plugged in, you risk accidentally
bumping the power switch. Installing or removing a device with the power on is catastrophic to
most components (unless they are hot-swappable) and introduces the risk of system-wide
electrical damage, not to mention the unpleasant surprise of receiving an electric shock. Also,
many power supplies continue to supply low-voltage power to the motherboard, even when

130
 switched off. By implementing proper grounding measures and unplugging the server, you can
avoid a potential mishap.
• Use a grounding kit—these come in several forms. At the lowest end, a portable grounding kit uses
a wrist strap with an alligator clip that attaches to the server chassis as in Exhibit 5-1. (Be sure to
attach it to an unpainted surface for best contact.) This has the same effect as touching the chassis
to equalize electrical potential, except that it is not temporary because the connection is constant.

Exhibit 5-1: A portable grounding kit and mat

A hardware repair bench normally includes a grounded floor mat (also called a map) that connects to
earth ground at a nearby electrical outlet (this does not affect the operation of other devices plugged into
that outlet). Another mat on the bench top is also grounded—either to the floor mat or independently to
another electrical outlet. The user wrist strap connects to either mat or the server chassis (see Exhibit 5-
2). Because mats, the wrist strap, and the chassis are all grounded, they possess the same electrical
potential, eliminating the risk of ESD. The wrist strap usually includes a resistor designed to negate a high-
voltage electrical charge, in case the technician accidentally touches a highvoltage item such as internal
components in the power supply or the monitor. Both items retain high amounts of voltage even when
unplugged. Full-time repair facilities normally ground the entire workbench.

Exhibit 5-2: A workstation grounding kit

For more information on specific grounding kits, refer to the following Web sites (also see
www.3m.com/ehpd/esd_training for more about ESD in general):
131
 • Specialized Products Company at www.specialized.net
• 3M Corporation at www.3m.com/ehpd/workstation
• Jensen Tools at www.jensentools.com

Other tips for avoiding ESD include:

• Take it to the mat—As you install and remove components, be sure to place them on the mat. You
might be tempted to place the computer itself on the mat, but that is unnecessary, provided you
grounded it properly. Leave it off the mat so that you have some workspace.
• Handle with care—If you find yourself without ESD protection, be sure you handle loose adapter
cards by the metal bracket that attaches to the chassis. The internal ground circuitry of the card is
connected to the bracket, so touching the bracket prevents ESD from damaging the card
components. If the device does not have a bracket (a motherboard, for example), handle it by the
edges and try to avoid touching any of the items on the surface of the card. Do not touch the metal
edge of the adapter that goes inside the expansion slot, because even minor soiling from oils in
your hand can contribute to corrosion. Do not stack components.
• Bag it—Place loose components into ESD-resistant bags. Remove any other type of packing
material, such as Styrofoam, bubble wrap, or cellophane, because they tend to build and hold a
static charge.
• Take off the jewelry—Metal jewelry can conduct electricity. It is a good idea to remove any jewelry
from your hands and wrists (watch, rings, and bracelets) before working on the server. Also,
jewelry can catch onto components and wiring and hinder your dexterity.

Now that you’ve made all the preparations for a server upgrade and have ensured that you are protected
against ESD, you can proceed to upgrade server components.

Upgrading the processor

Before upgrading the processor, perform a few tasks to prevent serious problems or damage. Primarily,
you want to verify that the BIOS and motherboard support the processor and, if using SMP, that the new
processor is compatible with the existing processor.

Generally, it’s a good idea to keep your BIOS version as up-to-date as possible—this unit addresses this
issue later. Because the BIOS directly affect the communication between the processor and the rest of the
system, the importance of BIOS compatibility is obvious. Besides keeping up-to-date, also access the BIOS
settings and verify that it supports SMP. A PC server usually allows 4-way SMP unless a mezzanine board
or other motherboard modification allows you to expand to 8-way or greater SMP.

Next, check the motherboard to see if the proposed processor is compatible. The form factor of a given
processor might physically fit in several different motherboards, but that does not mean it is compatible.
Recall that a processor operates at multiples of the bus speed. This is one of the factors that limit the
available upgrade path in your server.

For example, you cannot replace a 700 MHz Pentium III with a 900 MHz Pentium III.

The 700 MHz Pentium III is designed to operate on a 100 MHz bus. However, the 900 MHz processor is
designed to operate on a 133 MHz bus, which will not allow the processor to function properly on a 100
MHz bus. Also, a given chipset might not be compatible with the proposed processor. To determine the
compatible processor upgrade path for the server, you could research the motherboard manufacturer’s

132
Web site. However, it is better to verify the upgrade path with the actual server vendor because they
might have integrated something else into the system that affects upgrade compatibility.

Adding another processor to an existing processor involves more than simply making sure that both
processors are the same speed. The processors should be identical in every way, including cache size, form
factor, and stepping. For example, there are at least eight different 700 MHz Pentium III Xeon processors.
While they all operate at 100 MHz, they vary in L2 cache size—either 1024 KB or 2048 KB— and the new
processor cache must match the existing processor cache size. When adding another processor, verify
that the new processor’s stepping (the processor version) matches that of the existing processor. As Intel
manufactures processors, minor problems, incompatibilities, or inaccuracies might be discovered from
time to time. While the chances that these flaws will negatively affect server operation or compatibility
are minimal, Intel usually corrects them when practical, so two processors of the same speed and cache
still might not be exactly the same. You can also look on the processor to find its specification number (or
S-spec)—an alphanumeric code that uniquely identifies each processor version and is more specific than
the processor stepping. Exhibit 5-3 shows an S-spec (the last item on line 3, SL4MF). Notice also that the
first line indicates 1000 MHz, 256 KB L2 cache, 133 MHz bus speed, and 1.7 V power.

Exhibit 5-3: The S-spec on this processor is SL4MF

Processor slots and sockets

The processor and its respective slot or socket appears in two primary formats. Sockets accommodate a
processor format known as the Pin Grid Array (PGA) processor, which can be a flat, thin ceramic device
with hundreds of gold pins on the bottom. These types of processors fit inside a motherboard socket
receptacle that accepts each of the pins on the processor.

The most recent implementation is the Staggered Pin Grid Array (SPGA), which staggers the pin
arrangement to squeeze more pins into the same space (see Exhibit 5-4). The PGA has two formats:
standard PGA and the flipped chip PGA (FCPGA), referring to the fact that the processor die is “flipped”
upside down on the die. The PGA format makes no difference in the actual installation except that FCPGA
includes a fan in addition to the heat sink.

133
 Exhibit 5-4: An SPGA processor

The slot format processor can be a larger device that stands upright inside a motherboard slot, similar to
adapter or memory slots (see Exhibit 5-5). This format is referred to as the Single Edge Contact Cartridge
(SECC). The Single Edge Contact Cartridge2 (SECC2) format is a similar form factor, except it exposes the
contacts at the bottom. The slot format processor includes a specially constructed plastic and metal
housing that often includes an on-board cooling fan. Processor manufacturers have flip-flopped over the
years regarding which format they use. Most processors use the socket format; however, Intel seems to
favor the SECC slot format for the Xeon at this point.

Whichever format is used, we will consider the housing and CPU collectively to be the processor.

Exhibit 5-5: A Pentium III Xeon processor in the SECC format

Inserting the processor, either AMD or Intel, into a socket is an easy matter. Look carefully at the pins on
the processor and match them to the socket on the motherboard. You will usually see a bevel that
prevents you from accidentally inserting the processor in the wrong orientation. Simply match the bevel
on the processor to the bevel on the socket. Before you insert the processor, lift up a lever next to the
socket. The lever is the lock that holds the processor in place, and the feature is known as zero insertion
force (ZIF) because when you insert the processor, gravity alone should be enough to seat the processor
into the socket (see Exhibit 5-6). Sometimes you might have to help gravity a little bit, but very little
pressure is required. If the processor does not seem to drop easily into the socket, do not force it. If you
do, you may be buying another new processor.
134
Slots are keyed so that, again, you can only insert the processor in one orientation.

However, be aware that it requires significantly more force to insert the processor into a slot, and the
retention mechanism has guiding slots that facilitate this (see Exhibit 5-7).

Use the retention mechanisms on either end of the processor to release it from the slot if you need to
remove it later, but sometimes that can be a challenge as well. Most current processors require a slot.

Exhibit 5-6: A ZIF socket uses a lever to lock the processor in place

Exhibit 5-7: The slot and its retention mechanism

Removing the slotted processor properly is a matter of experience and getting a feel for removing the
processor, which is usually seated very firmly in the slot. You have to balance the objective of removing
the processor with a prudent degree of restraint, and this might require some experience. A safer method
is to procure a processor extraction tool. Flotron (www.flotron.com) makes such a tool (see Exhibit 5-8).

135
 Exhibit 5-8: A processor extraction tool

Processor cooling
Processors can get very hot, and you must exercise care in cooling them properly using a heat sink—an
attachment to the processor that either dissipates heat passively, through aluminum cooling fins, or
actively, using a small cooling fan, usually in addition to cooling fins. When you purchase a “boxed”
processor, it includes a cooling solution from the manufacturer. If it is “bare,” then it has no cooling
solution and you must determine a way to cool it yourself. Processors with only passive cooling depend
upon airflow from the power supply and/or other cooling fans in the system. Use active cooling even when
only passive is required and even if the boxed processor only includes passive cooling. (Other server
components, particularly ribbon cables for hard drives, can block airflow. Be sure to route cables so as not
to impede airflow.)Socket processors usually have a fan mounted on top of the cooling fins. The short fan
power cable is usually sufficient to reach the motherboard power connection for the CPU, and is often
marked on the motherboard as “CPU FAN.” Attaching this type of heat sink to the CPU (see Exhibit 5-9)
typically uses a clip that you hook to a notch on one side of the socket. On the other side, you press down
on the clip until it hooks onto the notch on the other side of the socket. Exhibit 5-9 shows the aluminum
cooling fins, clip, fan, and thermal tape.

Exhibit 5-9: A heat sink

Exhibit 5-10 shows the installed heat sink, fan, and fan power connection to the motherboard. Inserting
the heat sink might take considerable force; take care that the heat sink is oriented and aligned properly.

136
 Exhibit 5-10: The installed heat sink, fan, and fan power connection

If you want to provide maximum cooling, you can buy third-party heat sinks and fans that are usually much
larger than those that come from the manufacturer and provide even better cooling. Normally, these are
no more than $50. Processors that go into slots, such as the SECC for the Pentium II/III or the SECC2 for
the Pentium III Xeon processor, use a similar cooling method to the socketed processors, except that the
form factor is rectangular and larger to accommodate the larger slotted processors.

Most heat sinks include a small amount of thermal tape located at the contact point between the bottom
of the heat sink and the surface of the processor. The purpose of the thermal tape is to act as a conductor
through which heat is transferred from the processor to the heat sink. Otherwise, there would be a narrow
gap of air between the processor and heat sink, and air by itself is not a good conductor. Many technicians
prefer to apply inexpensive thermal grease instead. A small amount of grease fills the gap and draws away
heat better than thermal tape.

Notifying the operating system

In Windows NT and Windows 2000, you must notify the operating system of a change from one processor
to multiple processors; otherwise, the operating system does not recognize or use the additional
processor(s). The steps to make this change in Windows 2000 are as follows:
1. Right-click My Computer and choose Manage.
2. Select Device Manager in the left tree pane.
3. Expand the Computer node in the right tree pane. The type of computer appears under Computer,
and is probably Standard PC or Advanced Configuration and Power Interface (ACPI) PC.
4. Whichever type of computer appears, right-click it and choose Properties.
5. Click the Driver tab of the Properties sheet.
6. Click the Update Driver button.
7. Click Next to skip the introduction of the Upgrade Device Driver Wizard.
8. Select Display a list of the known drivers for this device so that I can choose a specific driver, and
then click Next.
9. Select the Show all hardware of this device class option.
10. Select the multiprocessor option that matches your computer, and proceed to the end of the
wizard.

137
 Exhibit 5-11: Choose ACPI Multiprocessor PC or MPS Multiprocessor PC as it applies

Microsoft offers a series of utilities with the Windows NT 4.0 Resource Kit. Run the UPTOMP utility from
the command prompt, or reinstall the operating system to use multiple processors.

In current servers, it is uncommon to have to set a jumper to the correct voltage, CPU multiplier, and bus
speed, but check documentation first, especially for Socket 7 or Super 7. If the voltage is set improperly,
the processor might not function correctly or excessive voltage might damage the processor. Socket 370,
Slot 1, Slot 2, and Slot A each adjust the voltage automatically. However, BIOS settings might be available
to over clock the performance (not recommended for servers).

Upgrading memory
Memory module upgrades will usually be DIMMs. (SIMMs is rarely found on current servers, though older
workstations might still use them.) Fortunately, you cannot accidentally install the incorrect memory
technology because the memory units have different installation notches, pin count, or length. For
example, you cannot install a DIMM in an RDRAM slot. However, you will still need to verify that the
memory you are upgrading is of the proper speed and matches other memory already installed in the
system. Technically, you can mix different speeds of memory, although it is not a good practice.

For example, you can install 133 MHz SDRAM modules on a 100 MHz bus with existing 100 MHz modules.
However, performance of all modules will be limited to 100 MHz. You cannot mix memory speed within a
single SIMM memory bank.

Identifying memory
Identifying existing memory and the memory the motherboard supports is more time consuming than
actually placing the memory modules. Absent the documentation that came with the server, you can
identify memory modules by reading the actual chips on the module, counting the chips, and measuring
the length. You can read the numbers on the module’s memory chips to determine the speed in

138
nanoseconds (ns), which correlates to a manufactured speed. For example, if the number ends in –10,
then you have 10 ns speed designed for a 66 MHz bus.

Determine if an error correcting function such as parity or ECC is present in the module.

There are several ways to verify this, but the simplest and most foolproof method is to count the number
of chips on the module. If the number is evenly divisible by three, then you have either ECC or parity
memory. For example, nine chips on the module is evenly divisible by three (9/3 = 3), identifying the
module as ECC or parity. If the part number on each chip is the same, then you have ECC, which includes
the error correcting function in each chip. If one of the chips has a different part number, then you have
parity memory, because the chip that is different is solely responsible for the parity function on behalf of
all the memory chips on the module.

Identify a SIMM as a 72-pin module with a single notch in the bottom center, measuring 4.26 inches, or
108.2 mm (see Exhibit 5-12), and identify a 168-pin DIMM with two notches at the bottom, measuring
5.26 inches, or 133.8 mm (see Exhibit 5-13). The notches in the bottom are spaced slightly differently. The
left notch spacing defines the module as registered, buffered, or un-buffered, and the right notch spacing
defines the module voltage at 5.0 V or 3.3 V. Again, the notching makes it impossible to make a mistake
with buffering or voltages when installing the modules.

Exhibit 5-12: A 72-pin SIMM

Exhibit 5-13: A 168-pin DIMM

The next generation of SDRAM, the 184-pin DDR DIMM, is an important memory module with which you
should be familiar. High-end workstations and servers are increasingly requiring DDR SDRAM because of
its extremely high throughput (up to 2656 MBps). You physically insert this module into the slot in the
same way as the 168-pin DIMM; however, you can identify it with a single key notch at the bottom
indicating its voltage (2.5 V) and with two notches on either end. The module is the same length as a
standard SDRAM DIMM, measuring 5.256 inches, or 133.5 mm (see Exhibit 5-14).
139
 Exhibit 5-14: A 184-pin DDR SDRAM DIMM

Also look at the physical characteristics of the module. Recall that RDRAM RIMMs have immediately
identifiable metal heat spreaders covering the memory chips (see Exhibit 5- 15). Another verification of
RDRAM is if empty memory slots on the motherboard have a C-RIMM that completes the continuity,
permitting memory data to pass through each slot.

Exhibit 5-15: An RDRAM RIMM with heat spreaders pulled away

Installing memory
Installing memory is a straightforward matter. Both SIMMs and DIMMs have notching that prevents you
from installing them in the wrong slots. The trick is to make sure that you have fully seated the memory
into the slot. (Memory modules are particularly sensitive to ESD, so be sure to exercise appropriate
precautions.)

For a SIMM, insert the module at a 45-degree angle to the SIMM slot by pressing down firmly. A notch on
one end of the module matches a protrusion on the slot to prevent backward insertion. Tilt the module
toward the locking clips until they snap into place. Tabs on the locking clips should fit precisely into holes
on the SIMM (see Exhibit 5-16). If the holes do not appear to match up, you probably have not pressed
the module down far enough into the slot. To remove the SIMM, use your fingers to separate the locking
clips from the holes in the module while simultaneously tilting the module up and out.

140
 Exhibit 5-16: Inserting a SIMM into the slot

You are more likely to be installing a DIMM, which requires 90-degree, straight downward insertion into
the DIMM slot. However, you will not tilt the module forward like the older SIMM. Instead, locking ejector
tabs automatically clamp onto the module when the DIMM is fully seated, though you might also press
them into place to verify proper locking (see Exhibit 5-17). To remove the DIMM, press down on top of
both ejector tabs simultaneously, and the module should come out.

Exhibit 5-17: The DIMM is inserted with locking ejector tabs

Remember the following guidelines when installing memory:

• Match the manufacturer of existing memory when possible. Production variances are probably
slight, but might produce an unacceptable risk in mission critical servers. On desktop PCs, mixing
manufacturers is a more acceptable risk.
• Try to plan server purchases with as few modules as possible to maximize future upgrade
possibilities. For example, instead of two 64 MB DIMMs, install a single 128 MB DIMM.
• If the memory modules vary in size, install the largest modules (in megabytes) in the lowest
numbered slot for best performance.
• Verify that the operating system is capable of recognizing the memory you install. For example,
Windows NT 4.0 can utilize only 4 GB of memory regardless of motherboard capacity.
• Installing large quantities of memory (2–4 GB or more) might require special configuration of the
NOS. Check with the vendor to verify. For example, Windows 2000 Advanced Server allows you to
access large amounts of memory by adding the “/PAE” switch (enabling Physical Address
Extension) to the Boot.ini bootup file.
• Large quantities of memory might also require specific hardware compatibility.
141
 • For example, Windows 2000 Advanced Server can utilize PAE only with a Pentium Pro processor
or later, 4 GB or more of RAM, and an Intel 450 NX or compatible chipset or later.
• Although the current motherboard and BIOS will automatically detect new memory, verify this in
the BIOS to confirm proper seating of the memory.
• Most NOSs include a virtual memory feature that corresponds to the amount of physical memory
installed in the system. When you increase the amount of physical memory, also increase the
amount of virtual memory, which might be referred to as a swap file. However, in large memory
implementations, it might be impractical to create a 4 GB swap file that matches 4 GB of physical
memory. In this case, about 2 GB is usually acceptable.

Memory modules and sockets designed for low-cost appeal might use tin contacts. Many sources tell you
to avoid mixing gold modules with tin sockets (and vice versa) because contact between the two creates
an oxidization known as “fretting corrosion,” affecting good electrical contact and possibly creating all
kinds of instability and memory error problems. However, this issue is less of a problem than it used to be
because tin contacts are rarely produced, especially for server platforms. All industry standard DIMMs use
gold contacts.

5.3 UPDATING THE BIOS, UPGRADING POWER

Supplies and adapters

In years past, it was impossible to update the BIOS without first removing the BIOS, which might involve
soldering tools. Current PCs and servers offer flash BIOS, which means you can download the most recent
update from the vendor’s Web site and apply it to the server without replacing the BIOS.

To keep on top of the most recent updates, consider checking with the server vendor to see if they offer
a notification service that sends you an e-mail when a BIOS update (or other updates such as a driver) is
available. Dell, for example, offers such a service. Before updating the BIOS, read all available
documentation about the update (a Readme.txt file usually accompanies the BIOS update). This is
important to determine the purpose of the update and if it solves any problems you might be
experiencing. Update the BIOS as follows:
1. Download the BIOS update from the system vendor. Although major BIOS manufacturers such as
Phoenix Software, Award Software, and American Megatrends Inc. (AMI) make most BIOS found
in servers, you should not seek or use updates from the BIOS manufacturer. Server vendors work
extensively to tailor specific BIOS exactly for the server vendor’s motherboard.
2. Execute the downloaded file. Typically, this will copy all necessary flash BIOS files to a blank floppy
disk. Note: the boot disk should be as clean as possible. It should be a system disk containing only
basic boot files (IOS.SYS, MSDOS.SYS, and COMMAND.COM) but no memory management drivers
such as HIMEM.SYS.
3. Record current CMOS settings. Most BIOS updates either automatically reset the CMOS for you or
recommend that you manually reset the CMOS to default settings. You can use the recorded CMOS
settings to reconfigure the CMOS to your preferences after the update is complete. (Instead of
writing down each setting, consider using the Shift+PrtScn—Print Screen—keys to send the CMOS
configuration screen to a locally attached printer. You will have to perform a manual form feed on
laser printers to print the page.)
4. Boot from the flash BIOS disk. If the extracted files do not create a bootable system disk, use any
DOS or Windows 9x system to first format the disk as a system disk.

142
 5. The flash BIOS usually present a list of options for what you want to do. Select the option to update
the BIOS.
6. After the BIOS update is finished, manually reboot the computer if the updated BIOS do not do so
automatically.
7. Upon reboot, access the CMOS settings and reset the values back to default. Otherwise, the system
might not function correctly.
8. Reboot the server and again access the CMOS settings, entering in your preferred settings
recorded in Step 3.

Recovering the BIOS

If the BIOS is corrupt, you cannot boot the system at all—not even with a valid flash BIOS disk. In fact, the
display adapter is probably unavailable, further complicating matters. Corrupt BIOS can be caused by a
number of things, including ESD to the BIOS EPROM, an interrupted flash BIOS update, or a virus.
Regardless of what causes the corruption, the system cannot function until you repair the BIOS. While
some systems might vary (specific instructions are probably in the documentation), the following are the
usual steps to take in recovering the BIOS:
1. Turn off the system power.
2. Having removed the server cover, look for a motherboard jumper that allows you to enter recovery
mode. Check motherboard documentation to find the jumper if it is not printed obviously on the
motherboard.
3. Insert the latest flash BIOS update into the floppy drive.
4. Turn on the system power. Corrupt BIOS usually means that even the video display is not
functioning, so you will have to listen to beep(s) to track what is taking place in the BIOS recovery:
o The BIOS sounds a single beep when it passes control to DOS on the bootable flash BIOS
floppy. DOS executes the Autoexec.bat file on the floppy, which in turn runs the flash BIOS
update executable.
o A single beep indicates commencement of the flash operation.
o Two beeps indicate a recovered system BIOS. o Two more beeps indicate successful
completion of the recovery.
o A constant series of beeps indicates a failed recovery attempt.
5. The recovery process usually takes between three to five minutes.
6. Remove the recovery diskette, and turn off server power.
7. Restore the motherboard jumper to its original position. (The server might have a read-only
jumper position that prevents viruses from writing to the BIOS.)
8. Turn on the system power, and access the BIOS settings to enter your preferences.

After rebooting, a CMOS checksum error or some other problem might appear. Try to reboot again (by
powering off and on) to see if that resolves the problem. If not, enter the CMOS setup utility to check and
save settings. You can often resolve CMOS checksum errors by accessing CMOS settings, saving the
settings, and then rebooting.

Upgrading a power supply

If you suspect a power supply is not performing reliably or to specifications, you can first verify this using
a digital multimeter (DMM). The DMM measures AC voltage, DC voltage, continuity, or electrical
resistance (see Exhibit 5-18). In this section, we are mostly concerned with the DC voltage readings.

143
 Exhibit 5-18: A digital multimeter

Memory parity check error messages are often an indicator of a problematic power supply. Recall that
memory must be continually charged to retain its data. If the power to memory fluctuates, you are likely
to lose data. If parity check messages consistently identify the same location in memory, then the problem
is probably with a bad memory module. If the parity check messages are in different locations, then the
problem is probably power-related.

Using a multimeter
A multimeter should help you clarify whether a power supply is operating within specifications. Although
a multimeter can be analog (using a needle to show measurements), you should use a digital multimeter
for best accuracy. To check the operating voltage of a power supply using a DMM, use the following
procedure:
1. Take the server cover off and power up the system.
2. The DMM has two probes: red and black. Find the power supply connector that connects to the
motherboard, and locate the Power Good pin (pin 8; third pin from the left on the unnotched side).
Insert the red probe into the connector at pin 8 (see Exhibit 5-19). Inserting the probe alongside a
live connection like this is known as backprobing.

Exhibit 5-19: Back probing the ATX power supply connector

3. Touch the black probe to a ground, such as the chassis.

4. The DMM should read between +3 V and +6 V. If not, then the system cannot see the Power Good
signal and does not start or run correctly.

144
 5. Repeat this process for other connectors (use the following tables as a reference). Connectors are
in the +/-3.3 V, +/-5.0 V, and +/-12.0 V range. You should not see more than 10% variance from
this range, and only 5% variance is acceptable for high-quality power supplies.

Replacing the power supply

Wise server choices include redundant power supplies, so hopefully replacing a power supply, while
urgent, does not involve downing the server. Many redundant power supplies include lights, beeps, or
both to alert you of an impending failure as well as total failure. Naturally, you want to replace the power
supply when warning of impending failure occurs, instead of waiting for total failure. Check with the
vendor documentation for specific instructions on replacing the power supply (PSU). Generally, replacing
a hot-swappable redundant power supply is quick and easy:
1. Verify that the replacement PSU is compatible. For hot-swappable server PSUs, use the server
vendor PSU instead of a third-party PSU for best reliability.
2. If necessary, unscrew the power supply from the chassis. Normally, redundant power supplies use
a lever or handle instead of screws to attach to the chassis. Flip the lever/handle or unscrew, and
then pull the PSU out of the server (see Exhibit 5-20). You will probably perform this action from
the back of the server in most cases, so make sure you have enough space to either slide the server
out of the rack far enough to reach behind the server from the front of the rack or that sufficient
space exists behind the server to both open a door (if present) and remove the power supply from
the back of the rack. Be sure to use both hands when removing the PSU; it might be heavy.

Exhibit 5-20: Pull down the handle and pull out the hot-swappable PSU

3. Replace the power supply by reversing the steps above. Hot-swappable PSUs plug directly into
connections on the server without requiring you to remove the cover and attach power cables
individually.

If the replacement power supply is an upgrade, not a matter of replacing a failed PSU, then you can
perform the above steps one PSU at a time until all are up to the new level.

Be sure to verify vendor documentation to see if the server supports the new power level.

For power supplies that are not hot-swappable, continued operation of the server is not possible when
the power supply fails, and replacing the PSU is an immediate concern. Do not attempt to repair the PSU;
even unplugged it retains a high level of dangerous electricity. It is more prudent to spend the money on
145
a new PSU. Also, if it suits you, there is more flexibility in choosing a different vendor with non-hot-
swappable power supplies. Make sure that the new power supply will fit the chassis. Some server vendors
use specially designed PSU form factors that prevent you from choosing a generic replacement.

Replace the non-hot-swappable PSU as follows:

1. Power down the system if it is running.
2. Unplug the power cable.
3. Remove the server cover. (At your discretion, consider not grounding yourself in this rare case. If
there is a chance that the PSU might ground to you, extremely high voltage could be discharged,
resulting in serious injury. Just keep touching the chassis and be extra careful about not touching
other ESD-sensitive components.)
4. Unplug all PSU cable connections to the motherboard, fans, drives, and so on.
5. Unscrew the PSU from the chassis. Non-hot-swappable PSUs do not have handles or levers to
release them from the chassis. Some servers might require that you first remove other
components such as adapters or hard disks in order to reach the PSU.
6. Remove the PSU. (If you have time, this is a good time to blow out or vacuum out dust.)
7. Replace with a new PSU by reversing the steps above. (Older AT power supply connections use
two connectors to connect to the power supply. If you see this, make note of the connector
orientation, because mixing them up can burn up the motherboard or components. Also, it’s easy
to accidentally make the “off by one” connection where you skew the connector to the board by
one pin.

Remember that the two black ground wires in each connector should be adjacent to avoid a mix-up (see
Exhibit 5-21). ATX-style motherboards, on the other hand, use a single connector that is keyed to prevent
backward installation.)

Exhibit 5-21: Older AT power supply connections

Upgrading adapters
Upgrading adapters is a fairly straightforward process. Actually, you cannot normally upgrade adapters—
any features or characteristics of the adapter are static except for driver updates, BIOS updates (as is the
case with SCSI adapters), video adapters with upgradeable memory, or LAN adapters to which you can
add a “wake on” LAN chip for network booting. To install new features, simply replace the adapter. When
installing the adapter into a slot, the following guidelines define what you can do:
• An 8-bit ISA card can fit into a 16-bit ISA or EISA slot.
A 16-bit ISA card can fit into a 16-bit ISA or EISA

146
 slot. A 32-bit PCI card can fit into a 32-bit or 64-bit
PCI slot.
• A 64-bit PCI card only fits into a 64-bit PCI slot.

When installing the adapter, apply firm, even pressure when guiding the adapter into the slot. Be sure to
orient the server so that you apply downward pressure (toward the tabletop) instead of sideways to avoid
tipping the server over. Try to visually line up the card with the slot to ensure success. Often, you will have
to remove adjacent cards first for better visibility and more working room. As you press down on the
adapter, you should feel the card “sink” into the slot (see Exhibit 5-22). If the retaining bracket is not flush
against the chassis, then the card is probably not fully seated. Once seated, screw in the retaining screw.
Be sure to cover empty slot openings on the back of the chassis to help provide good airflow.

Exhibit 5-22: Inserting an adapter into its slot

Though unusual with quality components, some cards cannot be fully seated because the retaining
bracket is flush against the chassis, preventing further downward motion.

If this is the case, you should choose a different adapter. Otherwise, you might have to use a pair of needle-
nose pliers and carefully bend the adapter to make it fit.

The AGP graphics slot is further away from the fastening point on the chassis than other bus interfaces,
such as PCI. As a result, AGP cards are a little more susceptible to “walking” out of the slot during
shipment. When you receive a new server, press down on all removable components, including the AGP
adapter, to ensure proper seating of cards. New motherboards often include an AGP card retention
mechanism that snaps over the slot and locks the card into place via a retention notch.

Upgrading adapter drivers is normally a matter of connecting to the vendor Web site, downloading the
drivers, extracting them to a temporary location on the hard drive, and installing them.

Upgrading the UPS

When upgrading the UPS, realize that the upgrade requires you to power down all systems connected to
the UPS, because an “upgrade” is really a replacement. For example, if a rack has three 8-way servers in
it and a large UPS for redundant power and the administrator knows that the UPS will not supply adequate
power after adding another 8-way server, he or she might proceed as follows:
1. Power off all load equipment attached to the UPS.
2. Replace the UPS with a model having sufficient power capacity and runtime.
147
 3. Plug in all equipment and again power up the systems on the new UPS.

Of course, because the servers must be offline while the UPS is being replaced, redundant servers must
perform the same services. If this redundancy is not available, notify the users of the planned downtime
and perform the upgrade during low usage periods. Also, perform a backup prior to the upgrade. The
batteries on a new UPS probably do not hold enough charge to provide adequate runtime if utility power
were to fail soon after the upgrade and a recent backup will assist the recovery. Alternatively, if the rack
has N+1 UPS redundancy, then the administrator can upgrade one UPS while another continues to provide
redundancy. If utility power is interrupted during the upgrade, the redundant UPS will continue to supply
power.

Server upgrade checklist

The following checklist should help you to perform carefully planned upgrades:
• Set a baseline of acceptable performance.
• Confirm NOS compatibility.
• Notify users if the upgrade affects them.
• Record settings if applicable (e.g., CMOS settings prior to BIOS upgrade).
• Verify available resources for the device (IRQ, DMA, I/O).
• Download the most recent driver(s) and BIOS upgrade.
• Read instructions, FAQs, and newsgroups.
• Inventory delivered parts.
• Perform a tape backup.
• Upgrade to the latest BIOS version.
• Perform the physical installation using sound ESD practices.
• Test and pilot the implementation.
• Reset the baseline.

Review Questions
1. When should you upgrade the server?
2. Why should you notify users of a planned upgrade?
3. What is an IRQ?
4. Why is a pilot program useful?
5. Before an upgrade, why would you perform a full backup instead of depending upon the normal
tape rotation?
6. Absent any grounding equipment, what can you do to protect against ESD?
7. Why should you keep your BIOS version updated?
8. What is a SECC?
9. Why are thermal tapes or grease sometimes used during cooling fan installations?
10. Open your computer’s case and look at the memory module(s) installed on the motherboard.
Using the information presented in this chapter, identify what kind of memory it is—SIMM, DIMM,
or RIMM; include information on parity, ECC, and speed where relevant.
a) Remove the memory module and reinstall it.
b) Start the PC. If the system does not POST, the module is not fully seated in the socket. Turn off the
power and try again.
11. What should you do prior to a flash BIOS upgrade?
12. What does a digital multimeter (DMM) measure?
13. Hot-swappable PSUs are typically identified by what feature?

148
 TOPIC 6
_____________________________________________________________

6. PLAN AND IMPLEMENT SERVER AVAILABILITY AND SCALABILITY

_____________________________________________________________

LEARNING OUTCOMES
After studying this topic you should be able to:
• Describe availability and scalability and differentiate between server clustering and Network Load
Balancing.
• Implement server clustering.
• Describe the concepts involved in server clustering.
• Describe the Network Load Balancing concepts; implement Network Load Balancing, and installing
applications on an NLB cluster.

6.1 AVAILABILITY AND SCALABILITY IN WINDOWS

Server 2003
As the information systems of an organization change and grow, server availability and scalability are
critical.
• Availability is the percentage of time that servers are providing service on the network
• Scalability is the ability to expand the number of clients or data that a server can support

How you implement availability and scalability depends on whether your applications are stateful or
stateless.

Windows Server 2003 provides two clustering mechanisms to provide availability and scalability. Server
clusters are used to provide highly available services that can failover from one server to another. Network
Load Balancing (NLB) is used to scale out applications and provide highly available services.

Server availability
Each time a server fails and ceases to provide services; there is a cost to your organization. Some of these
costs, such as lost staff time, are immediately apparent.

Other costs, such as customer loss of confidence, are not easy to quantify. You must understand what the
cost of system failure is to decide what can reasonably be done to prevent outages. For example, if the
cost of a 1 hour outage is $100,000 and the solution to prevent such an outage is $10,000 then it is obvious
the solution is worth the cost.

Some of the things that can cause a server or service to fail are:
• Hardware failure
• Network failure
• Administrator mistakes
• Operating system crash
• Application crash
A wide variety of options are available to provide higher server availability. These include:
149
 • Implement redundant hardware components such as RAID5 for disks and multiple power supplies
• Implement an Uninterruptible Power Supply (UPS) for temporary power outages
• Implement a backup power generator for longer power outages
• Create redundant paths through the network to move data around failed network components
• Use standardized procedures to perform tasks so that administrators understand how tasks are to
be performed
• Document the configuration of your network and servers so administrators understand the
implications of their actions
• Patch operating systems and applications to prevent crashes
• Use Windows Server 2003 clustering to implement server clusters
• Use Windows Server 2003 clustering to implement Network Load Balancing

Scalability
As an organization grows, the information systems used by that organization must be able to scale to
handle the addition of more users and more data. Scalability can be accomplished by scaling up or scaling
out.
• Scaling up is adding capacity to a single server
• Scaling out is adding capacity by adding additional servers to perform the same task

Scaling up
Scaling up is the first type of scaling thought of by most network administrators. If a server is unable to
handle its exiting load they will add more RAM, add a faster disk subsystem, or add more processors. This
is an appropriate way to add additional capacity, however, scaling up can be both limiting and expensive.

As systems are scaled up incremental improvements become more expensive and less effective because
the hardware is progressively more specialized. For example, the cost to migrate from a single processor
server to a dual processor server may cost $1000 dollars or less, and add another 90% of processing
capacity to the existing capacity.

Moving from a dual processor server to a quad-processor server may cost an additional $5000 dollars, and
improve processing capacity only an additional 80%. Moving up to eight or more processors is even more
expensive for the additional processing capacity received.

Scaling out
Scaling out avoids the problems of scaling up by adding more commonly available hardware rather than
using specialized hardware. For example, if you have a busy Web server hosting multiple Web sites, you
can buy an additional Web server and move some Web sites to the new server.

Scaling out works well, when many services are running on a single server. You can buy a new server and
move some services to the new server, however, it becomes more difficult when a single service is running
on a server and you want to scale out.

If a single Web site is running on a server and the server is overwhelmed, you can buy an additional server
and place another copy of the Web site on the new server. Then half of the users will need to be told to
access the Web site on Server1, and half the users told to access the Web site on Server2. This is awkward
for both users and the administrator.

150
Round-robin DNS, which allows multiple a records for a single host name, can be used to spread the load
between multiple servers hosting the same Web site. Round-robin DNS is unable to account for a failed
server. If round-robin DNS is spreading the load between two servers and one server fails then half of the
Web site users will be directed to the failed server.

Network Load Balancing (NLB) is the best solution for scaling out a single application.

Network Load Balancing can distribute the load between multiple servers like round robin DNS, but is also
able to sense a failed server and stop directing clients to the failed server.

Stateful and stateless applications

Applications can be broadly separated in the stateful and stateless applications. Stateful applications
require the server to retain knowledge about the client accessing the server.

For instance a shopping cart application on a Web site may require the server to keep a list of what the
user has selected in the shopping cart in a local file, or in memory.

Stateful applications are not well suited to scaling out because it is difficult for application developers to
track client information when client access is spread across multiple servers. It is easier to scale up a
stateful application.

It is possible to scale out stateful applications by storing sessions state information in a database or file
available to all servers. In addition Network Load Balancing attempts to address stateful applications by
setting affinity where requests from one IP address are always sent to the same server.

A stateless application does not require the server to retain knowledge about the client accessing the
server. For example, a simple Web site where the same pages are served out to all clients with no dynamic
content is stateless.

Stateless applications are well suited to scaling out because each request from a client is treated
independently. Stateless applications can be scaled up, but since it is more cost effective to scale out,
scaling out is the optimal choice.

Server clusters
A server cluster is used to provide highly available services. A service runs on a single server and if the
service fails on that server it is moved to another server in the cluster. It does not matter whether the
application crashed due to poor programming or the server hardware failed, it is automatically moved to
a new server. Moving a service from one server to another is referred to as failover.

When a service is moved from one server to another, clients will often notice a brief interruption in service.
This is a much shorter interruption in service than would be experienced if a service had to be manually
migrated to a new server.

Server clusters also provide the ability to upgrade applications with minimal downtime.

When an application needs to be upgraded on Server1 it is migrated to Server2, then the application is
upgraded on Server1 and migrated back to Server1. Users will see a few moments of interrupted service

151
while it is migrated from one server to another, but much less downtime than if the service were
unavailable during the entire migration process.

Server clusters are not available in all versions of Windows Server 2003. Server clusters are available only
in Windows Server 2003, Enterprise Edition and Datacenter Edition.

Up to 8 nodes are supported in each server cluster.

In almost all circumstances server clusters must have a shared storage. This can be in the form of a shared
SCSI bus or a fibre channel based storage area network (SAN). A SAN is a network that multiple server and
storage devices can attach to where the servers store files and data on the shared storage devices. A SAN
can range from $10,000 to millions of dollars.

Notice that the acronym SAN can also stand for system area network, which allows sharing of resources
between multiple computers beyond just storage. System area networks are rare, and in most cases when
SAN is used, it refers to a storage area network.

Server clusters are most useful for applications that have been scaled up and need high availability.

To use server clusters for an application, the application must use TCP or UCP packets.

Other protocols such as IPX, NetBEUI, or AppleTalk are not supported.

Server cluster configurations

Windows Server 2003 supports many configurations. Four of the most common are:
• Active/passive. An active/passive server cluster consists of two nodes where one is hosting
services and the other does not. A node is a server in a cluster. When the active node fails, services
failover to the passive node. If both nodes have the same capacity, you are assured that client
performance will not be reduced.
• N+I failover. An N+I failover server cluster is a variation of an active/passive server cluster. In this
configuration N nodes in the server cluster are active running services, and I nodes are passive in
the server cluster ready to accept failed over services. All active nodes in the cluster are configured
to failover to the passive nodes. Other than the overall limit of 32 nodes there can be any number
of active nodes (N) and passive nodes (I).
• Active/active. An active/active server cluster consists of two or more nodes, all of which are active.
In this situation a service fails over to an active node that is already running other services, because
the remaining node is running a new service in addition to its original services, it may be
overloaded and client performance may suffer.
• Single node virtual server. A single node virtual server does not have failover capability. It consists
of only one node. This configuration is often used for server consolidation. When an older server
is retired its resources can be migrated to a virtual server with the same name and IP address.
Then resources continue to be accessible without reconfiguring clients, a single node, and virtual
server has only one node it is used only to virtualize server resources to facilitate management.

Cluster applications
All applications must have certain characteristics in order to take advantage of server clusters. These are:

152
 • The application must use an IP-based protocol
• The application must be able to specify where application data is located so that it can be stored
on shared storage
• Client applications must retry lost connections to reconnect after an application failover

Some applications are aware of server clusters and are referred to as cluster-aware applications. Cluster-
aware applications are able to use the clustering API. This allows cluster-aware applications to receive
status notifications and manage the cluster. In addition cluster-aware applications can be taken failed
over and taken offline without any data loss.

The following are cluster aware services and applications:

• Distributed File System (DFS)
• DHCP
• WINS
• Microsoft Exchange Server
• Microsoft SQL Server

Cluster-unaware applications have a risk of data loss when an application is taken offline because the
cluster service terminates the process instead of performing a clean shutdown. In addition, the cluster
service attempts to monitor registry key changes made by cluster-unaware applications and replicate
those changes to the shared storage so they are available to the new node that an application has failed
over to.

Network Load Balancing

Network Load Balancing (NLB) spreads application requests from clients among two or more servers.
When NLB is configured an application is installed on multiple servers using a virtual IP address. The virtual
IP address is an IP address assigned to represent the application. When clients access the virtual IP address
their requests are redirected to one of the servers hosting the application.

When requests to the virtual IP address are redirected, the load is spread between the servers. NLB can
spread network requests evenly between the servers hosting the application or spread network requests
based on a weighting that you can assign.

Weighting the network requests allows you to place a heavier workload on servers with greater processing
power and higher amounts of RAM.

After NLB is configured you can add a new server at any time. This allows any application that can be
supported using NLB to be scaled out up to 32 servers. An NLB cluster is limited to a maximum of 32
servers, and can be implemented on any edition of Windows Server 2003. If scalability beyond 32 servers
is required then multiple NLB clusters can be combined using round-robin DNS. Round-robin DNS is
configured with an A record for each NLB cluster. When the A record is resolved then requests are spread
between the different NLB clusters. Availability is also enhanced by NLB. NLB recognizes when a server
has failed and stops distributing client requests to the failed server. This is a significant enhancement over
round-robin DNS as a load balancing mechanism; however, NLB is not able to recognize a failed service or
application, only entire server failure. NLB is best suited to applications that are stateless. Then client
requests can be spread over all of the servers without losing information about client sessions. If client
sessions are required then the session state data must be stored on the client computers or in a shared
location that all of the application servers have access to.
153
To use NLB for an application, the application must use TCP or UCP packets. Other protocols such as IPX,
NetBEUI, or AppleTalk are not supported. Applications that require data stored on a local drive letter
cannot use NLB. NLB does not allow data to be moved to from one server to another. NLB is not suitable
for application such as Microsoft SQL Server or Microsoft Exchange.

An example of Windows Server 2003 clustering

Before you begin working with Windows Server 2003 clustering you must be sure that you understand
how it can be used in different situations. The following example illustrates how both server clusters and
NLB can be used for a Web-based application.

Arctic University currently has a cumbersome manual system for taking student registrations. Each year
student’s line up for hours in an attempt to get the classes they want, however, the system is reliable and
works every year.

The board of governors has discussed computerizing the registration in the past, but it was always
dismissed because of concerns about reliability. This year when the issue is raised again they have decided
to ask the IT department to explore the possibility of a computerized registration system.

After careful thought the IT department has proposed a Web-based application based on Microsoft
clustering, as you see in Exhibit 6-1.

Exhibit 6-1: Windows Server 2003 clustering solution

As you can see, the backend of the system is a server with Microsoft SQL Server to store all of the student
and course information. To ensure reliability it will be installed on two Windows Server 2003 Enterprise
Edition servers scaled up with quad-processors and 2 GB of RAM configured as a server cluster. The Web
portion of the application will run on two low cost Windows Server 2003 Web Edition servers with one
processor and 1 GB of RAM that has been load balanced. If these two Web servers are not sufficient then
the solution can be scaled out by adding additional low cost Web servers.
6.2 INSTALL AND CONFIGURE SERVER CLUSTERS

154
The Cluster service is automatically installed during a Windows Server 2003 installation. You don’t need
to add it through Add/Remove Programs like other services. To configure a server cluster, you use the
Cluster Administrator (cluadmin.exe), as shown in Exhibit 6-2. This tool can be used to create new server
clusters or add new nodes to an existing server cluster. Another tool you can use to configure a server
cluster is the command line utility, cluster.exe.

Exhibit 6-2: Cluster Administrator

Creating a new cluster

When you choose to create a new cluster, the New Server Cluster Wizard appears. The wizard first
prompts you for the domain where the server cluster is located and a name for the cluster, as shown in
Exhibit 6-3. All cluster nodes must be part of the same domain. The name for the cluster is a virtual server
name that is used to manage the cluster.

155
 Exhibit 6-3: Cluster name and domain

After configuring the cluster name and domain you will be asked for the name of the server that is hosting
in the new server cluster, as shown in Exhibit 6-4. In addition there is an advanced button that allows you
to choose a typical (full) configuration or advanced (minimum) configuration. Only use the advanced
(minimum) configuration if you require special features that are not available in the typical (full)
configuration.

Exhibit 6-4: Choosing the first node

After you select the server that is hosting the first node in the cluster, the New Server Cluster
Administrator analyzes the suitability of your server for hosting a cluster. When the analysis is finished
you are shown the results, as shown in Exhibit 6-5. Click View Log to see a detailed log in text format.

Exhibit 6-5: Cluster analysis

156
After the analysis is complete you are asked for an IP address. This IP address is used for the virtual server.
It must be unique from any other IP address in the network. Clients access cluster resources through this
IP address and you can manage the cluster through this IP address.

Next you are asked for a cluster service account, as shown in Exhibit 6-6. This account is used by the cluster
service to logon. To run properly it needs to be assigned local administrative rights on all nodes in the
server cluster. This is done automatically by the New Cluster Server Wizard. In most cases the
Administrator account should not be used as the cluster service account because then your cluster can
fail when the Administrator password is changed, however, this is acceptable for testing purposes.

Exhibit 6-6: Cluster service account

The next screen is a summary of the installation options, as shown in Exhibit 6-7. You can review these to
ensure that you selected the correct options. There is also a Quorum button that you can use to modify
the quorum location. After the cluster installation is completed the current state of your cluster is shown
in Cluster Administrator.

Exhibit 6-7: Installation summary

157
How to Installing a single node virtual server cluster
1. Click Start¸ and then choose Run… Enter cluadmin Click OK
2. From the Action list, select Create new cluster Click OK
3. Click Next
4. In the Domain box, verify that Arctic.local is selected
5. In the Cluster name box, enter cluster<xx> Click Next
6. In the Computer name box, enter student<xx> Click Next
7. Expand the sections with error messages Click Next
8. In the IP Address box, enter 192.168.1.2<xx> Click Next
9. In the User name box, enter Administrator In the Password box, enter Password! Click Next
10. Click Next
11. Click Next Click Finish
12. Select Groups
13. Select Resources
14. Double-click Cluster Configuration Select Networks
15. Close Cluster Administrator

Why installing a single node virtual server cluster

1. To start Cluster Administrator.
2. To open the New Server Cluster Wizard.
3. To begin the New Server Cluster Wizard.
4. Where <xx> is your student number.
5. (If necessary.) Where <xx> is your student number.
6. The New Cluster Server Wizard will now analyze whether a cluster is possible. The analysis of your
server will generate warning messages.
7. (To read the messages.) The errors are because you do not have a shared storage system on your
server. The error also indicates that a local quorum will be configured. This is normal for a single
node virtual server cluster. Click Next To continue.
8. Where <xx> is your student number.
9. (To accept the current configuration.) The cluster will begin installation. After the cluster node is
installed a log of the installation process is presented to you
10. Notice the Cluster Administrator now shows the status of your new cluster.
11. To view the resource groups that has been created on this server.
12. To view the resources that has been created on this server. Notice that three resources have been
created: Local Quorum, Cluster IP Address, and Cluster Name.
13. To expand it.
14. To view the networks this cluster server is aware of.

6.3 SERVER CLUSTER CONCEPTS

A server cluster is used to provide high availability. To understand the best way to implement a server
cluster to protect your applications you must understand various concepts about server clusters. The
concepts include:
• Shared disks
• Quorum resource
• Cluster communication

158
 • Resource groups
• Failover and failback
• Virtual servers

All of these topics are covered in the following sections.

Shared disks
Shared disks are required for most server clusters. A shared disk is storage that all nodes in a cluster can
access. Shared disks are not required when a geographically dispersed server cluster is configured with an
alternate data synchronization mechanism, or when a server clusters has a single server for testing
purposes. When shared disk are required you can use either a shared SCSI bus or a fiber channel SAN.
Note that the shared disks used by a cluster are also referred to as cluster disks.

Shared SCSI bus

A shared SCSI bus is configured by using an external SCSI storage device and connecting it to the external
SCSI connector of both servers, as shown in Exhibit 6-8.

For this configuration to work properly you must ensure that the SCSI card in each server and the SCSI
drives support this configuration. Many SCSI cards and drives do not.

Exhibit 6-8: Shared SCSI bus

159
SCSI configuration requirements include:
• A SCSI card that can disable autobus reset
• All SCSI devices on the SCSI must have unique SCSI IDs. This includes the two
• SCSI cards
• Proper termination on the SCSI bus, the most effective mechanism for termination is external
Ycables with terminators
• SCSI hard drives that are multi-initiator enabled to support multiple SCSI cards on the bus.

A shared SCSI bus is the most inexpensive form of shared storage that can be used by server clusters. It
can often be configured for only a few thousand dollars. It is also the most difficult to configure because
there are often termination problems.

Fibre channel
Fibre channel is designed to be used for storage area networks. Consequently it is easier to configure and
faster than a shared SCSI bus, however, it is also much more expensive. For example, fibre channel host
bus adapters (HBA) can cost over $1000 whereas SCSI cards cost approximately $250. Exhibit 6-9 shows a
fibre channel SAN.

Exhibit 6-9: Fibre channel SAN

Requirement for fibre channel shared storage are:

• A fibre channel card for each node in the server cluster
• An external storage array that supports fibre channel
• A fibre switch to connect the nodes to the external storage array

160
Quorum resource
The quorum resource is a special disk resource used by the cluster service to store configuration
information and to arbitrate which node owns the cluster. The location of the quorum resource is specified
when a server cluster is created.

Like other resources, only one node at a time is allowed to own the quorum resource.

The first node in the cluster that is available becomes the owner of the quorum. When new nodes come
online they will receive configuration updates from active nodes in the server cluster. Only the first node
to come online receives updates from the quorum resource.

If there is a failure in network communication, the quorum resource is used to arbitrate which node keeps
its resources running. For example, in a two node cluster if network communication fails, the node with
the quorum will bring all resources online, and the node without the quorum will take all resources offline.

Cluster communication
Nodes in a server cluster are in constant communication with each other using heartbeat packets.
Heartbeats packets monitor which nodes in the server cluster are still up and available. These packets can
be UDP unicast packets or multicast packets.

Each network card in a node is designated as attached to a public, private or mixed network. Public
networks are used only for client to node communication and are never used for heartbeat packets.
Private networks are used only for node to node communication. Mixed networks can be used for client
to node, or node to node communication.

It is recommended that all nodes have at least two network cards. One of those network cards should be
configured as part of a private network, and the other part of a mixed network. When configured this way,
if the private network fails, then node to node communication can still be performed on the mixed
network.

Resource groups
Control of services and applications in a server cluster is based on resource groups. A resource group is a
logical grouping of all the resources that are required for an application or service to run. Resources can
be disk partitions, IP addresses, printers, services, and applications.

You can configure dependencies between resources in a resource group. For example, you can define that
an application resource is dependent on the IP address resource, and a disk resource. This way the
application will not be started until the IP address and disk are available.

Notice that dependencies can only be created between resources in a resource group.

They cannot be created between resource groups or between resources in different resource groups.
Resources can be in the following states:
• Online. The resource is operational and can be used by clients
• Offline. The resource is not operational and cannot be used by clients
• Online pending. The resource is being put online
• Offline pending. The resource is being taken offline
• Failed. The cluster service was unable to bring the resource online
161
Failover and failback
When a node or resource in a server cluster fails the failover occurs. If a server cluster node fails then the
resource groups on that node automatically failover to another node.

If a resource fails then the cluster service will attempt to restart the service. If the resource cannot be
restarted then the resource group it is a part of wills failover.

When failover occurs in a server cluster with more than two nodes a path must be defined for failover. If
all nodes in a server cluster are active then you should configure resource groups to failover evenly to
multiple nodes rather than overloading one node with all of the resource groups from the failed node.
You get to choose which nodes resource groups failover to.

After failover has occurred, failback is possible. Failback occurs when the original node hosting a resource
group is available again and the resource group is moved back to the original node. You can define
whether a resource group wills failback automatically, manually, or not at all. In a server cluster where all
nodes are active, failback is normally configured to be automatic so the load on each node can be reduced.
In a server cluster where passive nodes are available, the load generated on nodes by resources groups
that failover is not a concern and failback is normally a manual process if it is used at all.

Virtual servers
Resources in server clusters are always accessed through a virtual server. To client computers, this virtual
server appears to be a normal Windows Server 2003 server. It is really a collection of resources in a
resource group presented by the cluster service. The node that owns the resource group responds on its
behalf. When failover occurs, responsibility for that virtual server is passed to another node.

The virtual servers used by Windows Server 2003 clustering have improved integration with Active
Directory over Windows 2000 Server. Virtual servers are now published in Active Directory as a computer
object. This allows Kerberos authentication to be used, and browsing for the computer in a NetBIOS free
environment. Windows 2000 Server clustering requires NetBIOS for browsing, and used NTLM
authentication instead of Kerberos authentication.

Adding a printer to a virtual server

1. Click Start, and then choose Run…
You have already created a virtual server named cluster<xx>, where <xx> is your student number.
Now you’ll create a printer so that printing can be available on the virtual server.
Enter cluadmin
Click OK - To start Cluster Administrator.
2. Right-click Resources Choose New, Resource
3. In the Name box, enter Printers From the Resource type list, select Print Spooler
Click Next In the Possible Owners dialog box your server is listed in the Possible owners box.
4. Click Next To continue.
5. From the Available resources list, double-click Cluster IP Address Double-click Cluster Name
Double-click Local Quorum
Click Next - This makes all the Printers resource dependent on all of these resources. It will not be
brought online if all three are not available. This prevents the cluster from trying to provide
resources when it is not possible and ensures proper failover.
6. Click Finish - To accept the default Print Spooler Parameters.
162
 7. Click OK (To close the message box informing you of the successful creation.) Notice that your
resource named Printers is offline.
8. Right-click Printers- To start the Printers resource.
Choose Bring Online The status of the Printers resource is now online.
This is the same process you would perform if a resource failed and you wanted to bring it back
online.
9. Close Cluster Administrator
10. Click Start, and then choose Run…
Enter \\cluster<xx> Where <xx> is your student number. Click
OK To open the \\cluster<xx> window.
11. Double-click Printers and Faxes Double-click Add Printer
12. Click Next
Select Create a new port
Click Next (To trigger the Add Standard TCP/IP Port Wizard.) All cluster printers must be IP based.
13. Click Next
In the Printer name or IP address box, enter 192.168.1.15
Click Next There is no printer with the IP address 192.168.1.15 in the classroom. This IP address is
entered to simulate a network printer.
14. Click Next (To accept the default setting of Generic Network Card.) If a real printer were available
on the network you may not see this screen because the type of network device can be auto
detected.
15. Click Finish
16. Click Next To accept the default printer.
17. Select Keep existing driver (recommended) If necessary.
Click Next
18. In the Printer name box, enter ClusterPrinter Click
Next
19. Select Share name If necessary. Click Next
20. On the Location and Comment screen, click Next 21.
On the Print Test Page screen, select No If necessary.
Click Next
Click Finish
22. Close the Printers and Faxes on clusterxx window
23. Click Start, and then choose Run…
Enter \\cluster<xx> Where <xx> is your student number.
Click OK Notice that the shared printer you created on the cluster is now available and shared as
ClusterP.
24. Close the \\cluster<xx> window

Removing a server cluster node

1. Click Start, and then choose Run…
You’ll remove the final node from a server cluster. This is required before continuing on to perform
other activities using Network Load Balancing. NLB should not be configured on the same server
as server clustering or both will be unreliable.
Enter cluadmin
Click OK To start Cluster Administrator.
2. Right-click STUDENT<xx> Where <xx> is your student number. Choose Evict Node

163
 3. Click Yes To confirm evicting your server from the cluster.
4. Close Cluster Administrator

6.4 NETWORK LOAD BALANCING

An NLB cluster is used to scale out applications and services by making them seamlessly available to clients
on multiple servers at the same time. To effectively plan an NLB cluster you must understand a variety of
concepts. These are:
• NLB Driver
• Virtual IP address
• Application requirements
• Affinity
• Load balancing options
• Network communication
• Port rules

All of these topics are covered in the following sections.

NLB driver
The NLB driver is the software responsible for performing NLB on each host in an NLB cluster. It must be
enabled on each host in an NLB cluster. It is installed and available, but not enabled, on Windows Server
2003 by default, as shown in Exhibit 6-10.

Exhibit 6-10: Network Load balancing driver

164
The NLB driver operates between the network card driver and IP protocol. This allows it to intercept all
incoming IP traffic and filter it as defined by filter rules. It will also filter the traffic to ensure that only one
host in an NLB cluster responds to a request. Filtering traffic is required because all hosts in an NLB cluster
share a MAC address that is used for the cluster. All network cards using the cluster MAC address pass the
packets up to the NLB driver. The NLB driver then discards packets that are not relevant for the local host.

Enabling the NLB driver

1. Click Start, and then choose Control Panel¸ Network Connections You’ll
enable the NLB driver.
Right-click Classroom Choose Properties 2. Check Network Load Balancing
Click OK To close the Classroom Properties dialog box.

Virtual IP address
Each NLB cluster has a virtual IP address. Unlike a server cluster which is defined by a virtual server name
and IP address, an NLB cluster is defined only by an IP address. The virtual IP address must be unique on
the network. In addition, the virtual IP address must be on the same subnet as the hosts in the NLB cluster.
The virtual IP address must also be added as a secondary IP address to a network interface on the server.

Notice that the virtual IP address cannot be leased from a DHCP server. It must be statically assigned.

Application requirements
Applications that are suitable for NLB must meet certain requirements. These are:
• The application must use TCP or UDP
• Data modified by the clients must be synchronized between hosts in the NLB cluster or stored in a
central location
• If session state information is used then it must be stored on client computers, stored in a central
location, or affinity must be configured
• The application must not bind to a computer name
• The applications must not keep files open continually for writing

Affinity
Affinity is the process whereby responses to requests made originally to one host in an NLB cluster are
directed back to the original host. This is required for any applications that keep track of session state
information. This is also required if NLB is used for Terminal Services. Each client connecting to Terminal
Services must continue to communicate with the same server since their session exists only on one server.
SSL for secure Web applications can function without affinity but performance is must faster with affinity
enabled.

Affinity can be configured in three ways:

• None. Affinity is not performed; this is used when client’s applications are stateless
• Single. Affinity is performed based on the source IP address of the client, each request from the
same source IP address will be directed to the same host in the NLB cluster. This is the most
commonly selected option for stateful applications; this option is configured by default
• Class C. Affinity is performed based on the source network of the client. Each request from the
same Class C network is directed to the same host in the NLB cluster. This is used when clients are
accessing the application through multiple proxy servers with synchronized caching, and occurs
through some ISPs such as America Online (AOL).

165
Load balancing options
When hosts are part of an NLB cluster you can define how the load is balanced between them. There are
three options:
• Multiple Hosts – Equal
• Multiple Hosts – Load weight
• Single Host

When Multiple Hosts - Equal is chosen, the NLB cluster distributes the load evenly between all hosts in
the NLB cluster. This is best suited to situation where all hosts in the NLB have the same capacity for
processing. If one host is less powerful than others in the NLB cluster then the less powerful host will act
as a bottleneck for NLB cluster performance.

When Multiple Hosts – Load weight is selected, the NLB cluster distributes requests based on a load weight
value configured by you. The load weight value for each host can range from 0 to 100. The number of
requests directed to each host is based on its load weight value relative to other hosts. If two hosts have
the same load weight value, then each one receives the same number of requests. If the load weight value
of one host is twice that of another then the number of requests serviced by that host will also be twice
that high.

The percentage of overall requests serviced by a host is calculated by taking the load weight value assigned
to that host and dividing it by the total of the load weight values assigned to all of the hosts in the NLB
cluster. The following table shows an example of the percentage of load assigned to each server in an NLB
cluster with three hosts and the percentage of requests serviced by each.

Varying the load on each host in an NLB cluster is appropriate when varying levels of hardware are used.
In this way you can give less powerful servers a lower share of client requests.

When Single Host is selected the load is not spread between multiple servers. All requests to the NLB
cluster are directed to a single host. The host requests are directed to be selected based on the priority
setting for the host. Each host in an NLB cluster is assigned a priority setting between 1 and 32 with 1
having the highest priority. Single Host setting is used only to provide failover when a host fails. This option
is seldom used.

Network communication
When a client makes a request for a load balanced application, the request is sent to the virtual IP address
of the NLB cluster. Normally packets contain the MAC address of the network card in the destination
computer and that is how the destination computer understands the packet is addressed to it. In an NLB
cluster the packet needs to be addressed to all of the hosts in the NLB cluster, so, the MAC addresses of
the network cards in the NLB cluster hosts cannot be used.

All hosts in the NLB cluster share a MAC address. All client requests use the MAC address of the NLB cluster
as the destination MAC address. This allows all hosts in the NLB cluster to receive packets addressed to
the NLB cluster. Then the NLB driver loaded on each host accepts or discards the packet based on an
algorithm that takes into account affinity settings, load weighting, and priority.

166
NLB clusters can be configured to use either unicast MAC addresses or multicast MAC addresses. The
following sections discuss where each is appropriate. Both methods send all client requests to all hosts in
the NLB cluster.

Unicast
Unicast is the most common method used for distributing client requests. When this method is selected
all hosts in the NLB cluster use the same unicast MAC address. This unicast MAC address is used in place
of the MAC address that is embedded in the network card of each host.

If the NLB cluster hosts have only one network card they are unable to communicate among each other
when unicast is selected because they share a single MAC address. If the NLB cluster hosts have two
network cards then one network card can be used for receiving client requests and the other can be used
communication between NLB cluster hosts.

The source MAC address used by NLB cluster hosts is different from the MAC address used by the NLB
cluster. This is important to prevent a switch from learning the location of an NLB cluster host and mapping
it to MAC address used by the NLB cluster. Switches read the source MAC address in packets, and after
learning the location of a MAC address will direct all packets with that MAC address to only the port the
source MAC address was learned from. Since packets for the NLB cluster need to be delivered to all ports
on the switch this would be unacceptable.

A switch transmits a packet to all ports if it does not know the location of the MAC address. By preventing
a switch from learning the NLB cluster MAC address as a source MAC address, all packets addressed to
the NLB cluster will always be sent to all port on a switch. This does result in an overall increase in network
traffic because switches are not able to filter traffic.

Multicast
To prevent the inefficient use of switches, and allow NLB cluster hosts with a single network to
communicate amongst themselves, multicast MAC addresses can be used instead of unicast MAC
addresses. When multicast MAC addresses are used for the NLB cluster then the network cards still keep
their embedded unicast MAC addresses.

The source MAC addresses used by each host in the NLB cluster is the same multicast MAC address based
on the IP address of the NLB cluster. This is acceptable because switches will recognize this MAC address
as a multicast MAC address that can be assigned to multiple hosts.

If the switch supports Internet Group Management Protocol (IGMP) snooping it learns the locations of the
hosts in the NLB cluster and direct packets only to those ports to which an NLB cluster host is attached. If
the switch does not support IGMP snooping then it forwards the packets to all ports just as was done with
a unicast MAC address.

Port rules
Port rules are used to control what the NLB driver does with packets. When hosts in the NLB cluster receive
packets from clients port rules define what is done with each packet. Port rules are composed of a cluster
IP address, a port address range, a protocol, and a filtering mode which includes load weight and affinity.

If the cluster IP address is varied, different applications can be managed separately and use a different
filtering mode. For instance, the NLB cluster may have two Web-based applications, each using a different
167
IP address. The port rule for the first Web application can have one IP address, and the port rule for the
other Web-based application can have another IP address. Then the filtering mode including affinity can
be defined separately for each.

The port address range of a port rule can be used to define different filtering modes for applications using
the same IP address. For example, Terminal Services and a Web based application can run on the same IP
address in an NLB cluster. The port rule for Terminal Services affects only TCP port 3389, and the port rule
for the Web application affects only TCP port 80. Each of these rules can be used to set filtering modes
including affinity.

A default port rule always exists on an NLB cluster, and is sufficient in many situations.

The default port rule affects all clusters IP addresses, all ports, for TCP and UDP communication, and uses
the filtering mode Multiple Hosts; even load distribution, and single affinity. NLB clusters of VPN servers
or Terminal Services can use this default port rule if desired.

Implementing Network Load Balancing

To implement an NLB cluster, the NLB driver must be enabled on all servers that are to be hosts in the NLB
cluster. After this has been completed you can create a new NLB cluster using Network Load Balancing
Manager (nlbmgr.exe).

In Network Load Balancing Manager when you create a new cluster you will be presented with the Cluster
Parameters dialog box, as shown in Exhibit 6-11. In this dialog box you can configure the IP address, subnet
mask and Full Internet name (DNS name) of the cluster, the cluster operation mode, and allow remote
control. Allowing remote control means that the cluster can be managed remotely using the command
line utility nlb.exe. Allowing remote control is not recommended due to the security risk it presents.

168
 Exhibit 6-11: Cluster Parameters

The next dialog box, shown in Exhibit 6-12, lets you configure addition IP addresses that belong to the NLB
cluster. You add additional addresses if you want to bind individual applications to particular IP addresses
to manage them separately.

Exhibit 6-12: Cluster IP Addresses

After the Cluster IP Addresses have been configured you define the port rules. Only the default port rule
will be defined automatically during the installation process. Any other port rules must be created by you.
In this dialog box you can add new port rules or modify the default port rule. Editing the default port rule
is shown in Exhibit 6-13.

169
 Exhibit 6-13: Editing the default port rule

Next, you select an NLB cluster host to configure. On this screen you enter the IP address of a cluster host
you would like to configure, and the click Connect to get a list of interfaces, and then select the interface
you want to configure as part of the NLB cluster. The Host Parameters for the selected interface are
configured next, as shown in Exhibit 6-14. In this dialog box you set the Priority for the interface in the
cluster. This value must be unique among hosts in the NLB cluster. You can also define the Dedicated IP
configuration which is the IP address used for non-cluster communication. The Dedicated IP configuration
should be automatically filled in.

Exhibit 6-14: Host Parameters

You can also define the Initial host state for this host. The default selection is started which means that
when the server is started it will join the cluster. If you select Stopped then the host will not join the
cluster, you will need to manually start the host each time the server is restarted. If you select Suspended
then it will not join the cluster immediately after configuration but will on reboot. After the first host has
created the NLB cluster, others can join. When a new host is added to the cluster using Network Load

170
Balancing Manager, you first connect to the existing NLB cluster. Then you can add a new host from the
Cluster menu. When adding the new host you must supply its IP address, and select an interface for the
cluster address, as shown in Exhibit 6-15.

Exhibit 6-15: Adding a cluster host

After you have selected the host to add, you are asked to configure the host parameters exactly as you
are in the original cluster configuration. You do not get the chance to configure port rules because these
are taken from the already configured cluster. Now you can install applications on the NLB cluster.

Review Questions

1. Which term is used to describe the percentage of time a service is functioning correctly? a) Uptime
b) Scalability
c) Availability
d) Reliability
2. Which term is used to describe the ability to expand capacity of a server or application? a) Uptime
b) Scalability
c) Availability
d) Reliability
3. Which of the following can cause a server or service to fail? (Choose all that apply.) a) Hardware failure
b) Network failure
c) Administrator mistakes
d) Operating system crash
e) Applications crash
4. Which type of server cluster does not provide failover capability?
a) Active/passive server cluster
b) N+I failover server cluster
c) Active/active server cluster
d) Single node virtual server

171
5. Which type of server cluster will experience performance degradation if one node fails? (Choose all
that apply.)
a) Active/passive server cluster
b) N+I failover server cluster
c) Active/active server cluster
d) Single node virtual server
6. Which type of application in a server cluster can be migrated with no loss of data?
a) Cluster-unaware applications
b) Cluster-aware applications
c) Stateful applications
d) Concurrent applications
7. Which type of application is best suited to running on an NLB cluster? a) Stateless
b) Consecutive
c) Stateful
d) Concurrent
8. What is the maximum number of servers for an NLB cluster?
9. What edition of Windows Server 2003 is needed for NLB?
10. NLB allows data to be moved to from one server to another. True or false?
11. All resources in a resource group failover together? True or false.

12. When does failback occur? (Choose all that apply.)

a) Automatically when the original node becomes available
b) When the administrator manually performs it
c) Never
d) Automatically 10 minutes after the original node becomes available

172
 TOPIC 7
_____________________________________________________________

7. HARDWARE AND SOFTWARE MANAGEMENT

_____________________________________________________________

LEARNING OUTCOMES
After studying this topic you should be able to:
• Describe computer system hardware and the software needed to make it operate and perform
tasks.
• Describe how the operating system interacts with other software, such as application software,
the BIOS, and device drivers.
• Explain how the operating system loads, initializes, and starts application software.

7.1 HARDWARE MANAGEMENT

As shown in Exhibit 7-1, software controls hardware at the direction of a user. There are several layers of
software needed to make a computer system operate and perform tasks. You see these in Exhibit 7-1 as
application software, the operating system, and BIOS and device drivers.

Exhibit 7-1: Software needed to make computer hardware operate

Device drivers and BIOS are software written to interface with specific hardware devices. The BIOS (basic
input/output system) is permanently stored on microchips. The BIOS that the operating system uses to
help manage the system is called system BIOS and is stored on a BIOS chip on the motherboard. Device
drivers are stored on the hard drive until needed and are generally first added to the system as part of the
device installation process. Device drivers can come from a device manufacturer, or an operating system
can use its own device drivers, as shown in Exhibit 7-2.

173
 Exhibit 7-2: An OS relates to hardware by way of BIOS or device drivers

One advantage of using BIOS and device drivers as an interface with hardware is that it frees the operating
system and application software from having to know the specifics of how to communicate with a device.

For example, various printers understand data and commands according to various sets of rules and
standards called protocols. Application software and the operating system can pass print requests to the
printer driver, which communicates with the printer, as shown in Exhibit 7-3. With the device drivers doing
the interpreting, application software developers don’t have to include the specific protocol and
standards for every printer that might be used by the applications they write.

Exhibit 7-3: Application software and the OS passing print requests to the printer

The application software doesn’t even need to know which printer is being used, because Windows keeps
track of the default printer. The application sends print jobs to Windows for printing. Windows uses the
default printer, unless the user selects a different one from the Windows printer list. Windows knows
which device driver to call to execute the print job, because the device driver was assigned to that printer
when the printer was installed.

It's difficult to describe what an application, operating system, BIOS, and device drivers do unless you
know something about computer hardware. So first, you want to learn about several hardware
components common to most systems.

174
The motherboard
The motherboard, as shown in Exhibit 8-4, is the largest and most complex circuit board inside a computer
case. Because the CPU is central to all operations by hardware and software, all devices must somehow
connect to the CPU, and they do this by way of the motherboard. In order to allow for and manage this
communication, the motherboard has several slots, connections, embedded wires, microchips, and ports.

When you look on the back of a computer case, you can see some ports that are connected directly to the
motherboard. These provide a way for external components to connect to the CPU.

Exhibit 7-4: The motherboard is the largest circuit board inside a computer case

Exhibit 7-5 shows the ports provided to the outside of the case by this motherboard: a keyboard port, a
mouse port, two serial ports, two USB ports, and a parallel port. A serial port is called a serial port because
data travels serially, that is, one bit follows the next. This port is often used for an external modem or
serial mouse, a mouse that uses a serial port. A parallel port carries data in parallel and is most often used
by a printer. A USB (universal serial bus) port is a newer port used by many input/output devices, such as
a keyboard, printer, scanner, or mouse. You want to learn how the OS configures and manages each of
these ports and the devices that use them.

175
 Exhibit 7-5: Input/output devices connect to the computer case by ports

Listed below are the major components found on all motherboards, some of which are labeled in Exhibit
7-4.
• Central processing unit (CPU)
• Random access memory (RAM)
• Traces or wires
• Expansion slots
• ROM BIOS memory chip
• CMOS configuration chip
• Power supply connections

The CPU
The CPU is central to all processing done by a computer. Every command from the operating system is
passed to the CPU, which controls all the hardware components in the computer.

The CPU is installed in a slot or socket on the motherboard (refer back to Exhibit 7-4).

All x86 CPUs today operate in one of two modes: real mode and protected mode. When the CPU operates
in real mode, the CPU can address only up to approximately 1MB of RAM and doesn’t provide any features
to prevent errant applications from interfering with memory that belongs to another application. When

176
the CPU operates in protected mode, the CPU allows the operating system to use features that protect
one application from another.

Today, all CPUs for personal computers begin processing in real mode when they’re first turned on and
must be instructed to change over to protected mode. DOS stays in real mode, if the EMM386.exe isn’t
loaded. Other operating systems quickly switch to protected mode, which is much faster than real mode.

The speed of a CPU is partly determined by how much data it can process at one time and how much data
it can send or receive at one time. Every CPU has lines coming to it that are embedded on the motherboard
and collectively called a bus. These lines are devoted to various purposes. Some lines on the bus are
designated to carry data, and they’re called the data bus or data path.

Early CPUs used an 8-bit data path and processed 16 bits at a time. Today's CPUs can use one size for the
data path but another for internal processing. For example, the Pentium III CPU uses a 64-bit data path
coming to and from the CPU, but internally, it processes 32 bits at a time. Other lines on a bus are used
for addresses, control signals, and voltage.

Note: A new CPU, the Intel Itanium, operates in 64-bit mode. The 64-bit versions of Windows XP and
Windows Server 2003 Enterprise and Datacenter editions are designed to use this type of CPU.

Memory or RAM
PCs use RAM (random access memory) for short-term storage of data and programs.

RAM microchips are used to hold data and instructions temporarily while the CPU processes both. These
microchips are stored on tiny circuit boards called memory modules. Memory modules can be SIMMs,
DIMMs, or RIMMs. The most common module for today's motherboards is a DIMM. A memory module is
installed in memory slots on the motherboard designed to hold a particular type and speed of module.
Using Windows 9x, you can find out what type of CPU and how much memory you've installed by
rightclicking the My Computer icon on your desktop, choosing Properties from the shortcut menu, and
clicking the General tab.

Memory is useless to the system until it has been assigned addresses that the operating system, device
drivers, and BIOS can use to communicate with it. These memory addresses are numbers assigned to each
usable cell of memory, and the assignments are normally made when the OS is first loaded.

Buses on the motherboard

All hardware devices are directly or indirectly connected to the motherboard, because they’re all
dependent on the CPU on the motherboard for processing their data. Each bus provides a way for devices
to connect to the motherboard and, ultimately, to the CPU. Sometimes a bus provides a port on the
outside of the computer case for an external device to connect to by way of a cable or cord, or the bus
might provide a slot on the motherboard into which a circuit board fits. In any case, a device always
connects to a single bus on the motherboard. A motherboard can have several different buses; each type
of bus has data lines, address lines, and control lines. An operating system relates to the bus, so the OS
must support it in order for it to be used by the system.

Next you want to learn about some common buses and the expansion slots they support.

177
ISA bus
The first motherboards of the 1980s only had one bus, the system bus, which supported several ISA
(Industry Standard Architecture) slots. The first ISA slot had only eight lines for data and was called the 8-
bit ISA slot. It had 20 address lines and ran at 4.77 MHz, which means that data was transferred on the
bus at a rate of 4,770,000 transfers per second.

Later, the 8-bit ISA slot was improved by adding an additional 8 lines for data, and this version is known
as the 16-bit ISA slot. This slot runs at 8.33 MHz and has 24 lines for the address bus. All operating systems
for personal computers today support the 16-bit ISA slot, although its use is diminishing in favor of faster
I/O buses. An 8-bit expansion card, as shown in the exhibit, can use the first part of the 16-bit slot.

System buses
Today's system buses run up to 400 MHz and have a 64-bit wide data path and a 36-bit or wider address
path. These system buses, sometimes called memory buses, front-side buses, or local buses, always
connect directly to the CPU and memory, but are much too fast to support the slow ISA slots, which
connect to the system bus by way of slower I/O buses that act as intermediaries.

PCI bus
The PCI (Peripheral Component Interconnect) bus was invented for devices that are faster than the 16bit
ISA bus but can't run as fast as the system bus between memory and the CPU. The PCI bus runs in sync
with the system bus at one-third or one-half the speed. PCI uses a 32- or 64 bit path for data transfer. The
maximum throughput of 32- bit operation is 132 MBps, whereas maximum throughput for 64-bit
operation is 264 MBps. All Windows operating systems support PCI. In addition, the PCI bus is processor-
independent, which means that it can be used for many non-Intel systems, such as the Apple Macintosh.

AGP bus
The AGP (Accelerated Graphics Port) bus was designed to accommodate a fast video card, which is a circuit
board that controls the monitor. A motherboard has only a single AGP slot to accommodate one monitor.
Rather than using the PCI bus for graphics data, AGP introduces a dedicated point-to-point channel
between the graphics controller and the main memory. The AGP bus is based on the PCI specification,
which is 32 bits wide.

USB
USB (Universal Serial Bus) is designed to provide ports off the motherboard for slower external devices,
such as a mouse, keyboard, scanner, video camera, or digital telephone. You saw two USB ports on the
back of a computer case in Exhibit 7-5. The USB bus supports up to 127 devices, which can be daisy chained
together and connected to a single USB port on a motherboard. Windows 95 Service Release 1 has limited
support for USB. USB is fully supported in Windows 98, Me, 2000, XP, and 2003.

IEEE 1394 FireWire

IEEE 1394, also known as FireWire, is a new bus developed for high-bandwidth devices such as digital
cameras, digital camcorders, and videodisc players. IEEE 1394 adheres to the Win32 Driver Model (WDM)
by using the Open HCI standard. IEEE 1394 allows the direct connection of up to 63 devices. However, if
splitters, bridges, and repeaters are used, then IEEE1394 can connect up to 1,023 buses, which allow for
the connection of 64,000 devices. Windows 98/Me/NT/2000/XP/2003 support FireWire.

178
 Exhibit 7-7: Three ISA slots

Managing buses
Device Manager under Windows 9x and Windows 2000/XP/2003 (Windows NT doesn’t have a Device
Manager) is the primary tool used to manage bus resources and the devices that use them. To access
Device Manager using Windows 9x, right-click the My Computer icon on the desktop, choose Properties
from the shortcut menu, and click on the Device Manager tab. For Windows 2000, from the System
Properties window, click on the Hardware tab, and then click Device Manager. The Device Manager tab
of the System Properties window for Windows 98 is shown in Exhibit 7-8. You can see two PCI devices, a
modem, a network card, and the AGP video card listed in the window. The USB controller is also shown.

179
 Exhibit 7.8: The Device Manager tab for Windows 98

Motherboard system resources

Software has resources to control hardware, and hardware has resources to alert software that it needs
attention. Think of a system resource as a tool that either hardware or software uses to communicate
with the other. There are four types of system resources: memory addresses, I/O addresses, interrupt
request numbers (IRQs), and direct memory access (DMA) channels. The following table shows these
system resources used by software and hardware and defines each:

System resource Purpose

Memory addresses- Numbers that are assigned to physical memory located either in RAM or ROM chips.
Software can access this memory by using these addresses. On the motherboard, memory addresses are
transferred on the address bus.

IRQ -A control line of a motherboard bus that a hardware device can use to signal the CPU that the device
needs attention. Some lines have a higher priority for attention than others. Each IRQ line is assigned a
number (0 to 15) to identify it.

I/O Addresses -Numbers assigned to hardware devices that software uses to command a device. Each
device “listens” for these numbers and responds to the ones assigned to it.

On the motherboard, I/O addresses are transferred on the same address lines used by memory addresses.

180
DMA channel -A number designating a channel through which the device can pass data to memory
without involving the CPU. Think of a DMA channel as a shortcut for data moving to/from the device and
memory. Physically, DMA channels are lines on the motherboard.

System resource communications

Hardware devices signal the CPU for attention using an IRQ. The operating system addresses a device by
one of its I/O addresses. The OS looks at memory as a hardware device and addresses it with memory
addresses, and DMA channels are used to pass data back and forth between a hardware device and
memory. As you read about these system resources, you might be interested in examining your own
computer to see how the system resources are used. In Windows 98, go to Device Manager, right-click
Computer, and then choose Properties. The Computer Properties window is displayed. In Windows 2000,
go to Device Manager and, from the menu, click View, Resources by type.

Exhibit 7.9: Device manager shows the current assignments for system resources

Interrupt request number (IRQ)

When a hardware device needs the CPU to do something, such as when the keyboard needs the CPU to
process a keystroke after a key is pressed, the device needs a way to get the CPU's attention, and the CPU
must know what to do once its attention is turned to the device. These interruptions to the CPU are called
hardware interrupts, in which the device initiates an interrupt by placing voltage on the designated IRQ
line assigned to it. This voltage on the line serves as a signal to the CPU that the device has a request that
needs processing. Often, a hardware device that needs attention from the CPU is referred to as needing
servicing. Many processes that the CPU carries out are initiated by interrupts and are said to be interrupt-
driven. Software can also issue an interrupt to the CPU, so that the software can have access to a device.

Exhibit 7.9 shows IRQ 11, which is being shared by several devices on a Windows 2000 system. Look to
the left of the IRQ number and see that the ISA bus controller is managing some of these IRQs and that
the PCI bus controller manages the others. ISA managed IRQs can’t be shared, but PCI-managed IRQs can
be shared. Also notice that the secondary IDE channel uses IRQ 15, and the primary IDE channel uses IRQ
14.
181
IDE channels are used to connect hard drives, CD-ROM drives, Zip drives, and DVD drives to the
motherboard. With interrupts, the hardware device or the software initiates communication by sending
a signal to the CPU, but a device can be serviced in another way, called polling. With polling, software is
constantly running that has the CPU periodically check the hardware device to see if service is needed.
Not very many devices use polling as the method of communication; most hardware devices use
interrupts. A joystick is one example of a device that uses polling. Software that’s written to manage a
joystick has the CPU check the joystick periodically to see if the device has data to communicate, which is
why a joystick doesn’t need an IRQ to work.

I/O addresses
Another system resource that’s made available to hardware devices is input/output addresses, or I/O
addresses. I/O addresses, or port addresses, sometimes simply called ports, are numbers that the CPU can
use to access hardware devices. The address bus on the motherboard sometimes carries memory
addresses and sometimes carries I/O addresses. If the address bus is set to carry I/O addresses, then each
device is listening to this bus. If the address belongs to it, then it responds; otherwise it ignores the request
for information. In short, the CPU knows a hardware device as a group of I/O addresses. If it wants to
know the status of a printer or a floppy drive, for example, it passes a particular I/O address down the
address bus on the motherboard.

Exhibit 7.10: How I/O address lines on a bus work

Because IBM made many address assignments when the first PC was manufactured in the late 1970s,
common devices such as hard drives, floppy drives, and keyboards have no problem with I/O addresses.
Their controllers can simply be programmed to use these standard addresses. Devices such as scanners
or network cards that weren’t assigned I/O addresses in the original IBM list can be configured to use
more than one group of addresses, depending on how they’re set up during either the installation process
or during the startup (boot) process.

182
 Exhibit 7.11: Windows 98 I/O address assignments

COM and LPT assignments

In the previous table, notice the COM and LPT assignments. COM1 and COM2 are preconfigured
assignments that can be made to serial devices such as modems, and LPT1 and LPT2 are preconfigured
assignments that can be made to parallel devices such as printers. For example, rather than assigning an
IRQ and some I/O addresses to a modem, the modem is configured to use the assignments previously
made to COM2, making it easier to configure the modem and to avoid conflicts with other devices that
also need an IRQ and some I/O addresses. It’s common for a system to assign COM1 or COM2 to a serial
port and LPT1 or LPT2 to a parallel port, as shown in Exhibit 8-5. When an external modem connects to a
serial port that’s assigned COM1, the modem is using the system resources assigned to COM1. If a printer
is connected to the parallel port assigned to LPT1, it’s using the system resources assigned to LPT1.

Memory addresses
An operating system relates to memory as a long list of cells that it can use to hold data and instructions,
somewhat like a one-dimensional spreadsheet. Each memory location or cell is assigned a number
beginning with zero. These number assignments are made when the OS is first loaded and are called
memory addresses. Think of a memory address as a seat number in a theatre. Each seat is assigned a
number regardless if it has someone sitting in it. The person sitting in a seat can be data or instructions,
and the OS doesn’t refer to the person by name but only by the seat number. For example, the OS might
say, “I want to print the data in memory addresses 500 through 650.”

183
 Exhibit 7.12: Memory addresses are assigned like seats in a theatre

These addresses are most often displayed on the screen as hexadecimal (base 16 or hex) numbers in
segment/offset form. For example, C800:5 in decimal are 819,205. Let's first look at how memory
addresses are assigned and then turn our attention to how the CPU makes use of them.

Assigning memory addresses

Remember that the CPU uses lines coming to and from it to receive and send data. It has another group
of lines called the address bus that it uses to send and receive the memory addresses that it uses to access
memory. For example, if it wants to read the data in memory address 819,205, it must send the value
819,205 to memory to say to memory, “Send me the data stored in cell 819,205.” It does this by sending
the binary value for 819,205 on the address bus, which is 11001000000000000101. Each line on the
address bus must hold one of the bits for this number. In this case, 20 lines are required. Therefore, the
largest memory address the CPU can use is limited by the number of lines on the motherboard devoted
to the address bus.

Note: Windows offers a calculator that can quickly convert numbers in binary, digital, and hexadecimal.
You can use it to follow along with the conversions used here. Enter a number in one number system, and
then click another number system to make the conversion. To access the calculator in Windows 9x or
Windows NT/2000/XP, click Start, and then choose Programs, Accessories, and Calculator.

Early CPUs had only 20 lines on the bus available to handle addresses, so the largest memory address the
CPU could use was 11111111111111111111, which is 1,048,575 or 1,024K or 1 MB of memory.

DOS and applications use the first 640K of memory, while the BIOS and device drivers use the addresses
from 640K up to 1024K. Then newer CPUs and motherboards were developed with 24 address lines and
more, so that memory addresses above 1024K became available. This is called extended memory.
Windows 9x still uses these same divisions of memory, although it makes more use of extended memory.
Memory addresses are expressed using hexadecimal notation. Because the hex numbers in upper memory
begin with A through F, the divisions of upper memory are often referred to as the A range, B range, C
range, and so on, up to the F range. The previous table applies only to DOS and Windows 9x. Windows

184
NT/2000/XP/2003 use an altogether different memory-mapping design, in which there’s no conventional,
upper, or extended memory; it's all just memory.

Using Windows 9x Device Manager, see how the first 1 MB of memory addresses are assigned,. To view
the list, select Computer and click Properties, then click Memory. The system BIOS is assigned memory
addresses in the F range of upper memory. This F range is always reserved for motherboard BIOS and is
never requested by other programs. When the CPU is first turned on and needs a program to know how
to boot up, it begins with the instructions stored on the ROM BIOS chip that are assigned to these memory
addresses.

Exhibit 7.13: How the 1st megabyte of memory addresses is assigned

Using memory addresses

Once memory addresses have been assigned to memory, they can be used for communication with all
software layers. As you saw in Exhibit 7-2, BIOS, device drivers, OS, and application software are working
when a computer is running. During output operations, application software must pass information to the
OS, which in turn passes that information to a device driver or to the BIOS. The BIOS and device drivers
managing input devices must pass information to the OS, which passes it to the application software.

These layers of software all identify the data they want to share by referring to the memory address of
the data.

185
 Exhibit 7.14: Applications, the OS, and drivers passing data

Direct memory access (DMA) channels

Another system resource used by hardware and software is a DMA (direct memory access) channel, a
shortcut method whereby an I/O device can send data directly to memory, bypassing the CPU. Some
devices, such as hard drives or sound cards, are designed to use DMA channels. Others, such as the mouse,
are not. Those that use the channels might be able to use only a certain channel, say channel 3, and no
other. Alternately, the BIOS might have the option of changing a DMA channel number to avoid conflicts
with other devices. Conflicts occur when more than one device uses the same channel. DMA channels
aren’t as popular as they once were, because their design makes them slower than newer methods.
However, slower devices such as floppy drives, sound cards, and tape drives may still use DMA channels.
Notice that, in this Windows 98 system, a sound card uses DMA1 and a floppy drive uses DMA2. DMA3 is
available for a printer, and DMA4 is designated for access by the DMA controller for the higher DMA
channels.

Exhibit 7.15: DMA channel assignments

186
Using system resources
Now let's look at examples of how the system resources help carry out hardware and software interrupts.
This helps show how the resources work together.

Hardware interrupts
Once the CPU is interrupted by an IRQ, the job of the IRQ is over, and the CPU must handle the
interruption. To do that, the CPU must use a program designed to interface with the device. This program
to process can interrupt is called the request handler or interrupt handler. The program is either a part of
the total BIOS in a system or is a device driver that had previously been installed on the hard drive.

When a keyboard uses an IRQ to request attention from the CPU, the following steps are performed:
1. A key is pressed on the keyboard. The keyboard controller puts voltage on its assigned IRQ 1 to
the CPU, saying, “I need attention.” The CPU sees the IRQ, acknowledges it, and turns its attention
to servicing it. By sending the acknowledgment, it’s requesting that the device controller send a
number called an interrupt (abbreviated INT) that tells the CPU what service the device needs.
2. The keyboard controller sends INT 9 to the CPU, saying that it has a pressed key to process. The
CPU uses this value to locate the program to handle this interrupt.
3. The CPU looks to a table in RAM called the interrupt vector table, or vector table, that contains a
list of memory address locations of interrupt handlers. The INT value passed to the CPU by the
controller points to a row in the interrupt vector table. This row stores the memory address for
the instructions to service the keyboard, a portion of system BIOS.
4. The CPU looks to the location in memory of the request handler and begins to follow the
instructions there.

The CPU, following the interrupt handler instructions, processes the keystroke.
Note: The BIOS and operating system initialize the interrupt vector table during booting, but later another
program can modify the vector table to change the interrupt handler location to execute another program
instead. This is a common method that a virus uses to propagate itself.

Software interrupts
In the hardware interrupt example, two of the four system resources were used (memory addresses and
an IRQ). The keyboard controller used an IRQ to initiate communication.

When the software initiates communication, such as when the user of word processing software gives the
command to save a file to the hard drive, this is known as software interrupt. Both hardware and software
interrupts use the same numeric INT (interrupt) values to communicate their requests to the CPU. The
interrupt value for a call to the hard drive for I/O interaction is INT 13.

The software interrupt process, uses the following steps:

1. The CPU receives the software interrupt, INT 13.
2. The CPU turns to the interrupt vector table, using the INT value to locate the correct request
handler.
3. The CPU locates the handler in memory.
4. The CPU alerts the hard drive that instructions are forthcoming by sending one of the hard drive's
I/O addresses over the address bus.
5. The CPU follows the instructions of the request handler, in this case, system BIOS, to manage the
hard drive.

187
BIOS on the motherboard and other circuit boards
The BIOS is a hybrid of two worlds. It's technically both hardware and software, since it's the intersection
point of the two and must communicate with both. Because this software is permanently stored on a
microchip, it’s called firmware, and the chip is called a ROM (read-only memory) BIOS chip. The
motherboard and other hardware devices contain ROM BIOS chips. The motherboard contains a vital ROM
BIOS chip, that contains the programming necessary to start the computer, called startup BIOS. The
motherboard also contains other fundamental BIOS programs to control I/O devices, such as the floppy
disk drive and the keyboard, called system BIOS.

Advanced configuration and power interface (ACPI)

Some types of BIOS on the motherboard and operating systems support a power saving feature using the
standards developed by Intel, Microsoft, and Toshiba, called the Advanced Configuration and Power
Interface (ACPI) standards. Using ACPI, a system can be powered up by an external device, such as a
keyboard. Windows 9x and Windows 2000 support ACPI, as do most new motherboard BIOS. Microsoft
calls an ACPI-compliant BIOS a “good” BIOS. To see if your BIOS is ACPI compliant, check this Microsoft
URL: www.microsoft.com/windows2000/professional/howtobuy/upgrading/compat

Plug and play

Another feature of both the BIOS and the OS is Plug_and_play (PnP), a standard designed to make the
installation of new hardware devices easier. If the BIOS are Plug_and_play, then it begins the process of
configuring hardware devices in the system. It gathers information about the devices and then passes that
information to the operating system. If the operating system is also Plug_and_play compliant, it uses that
information to complete the hardware configuration process. For a system to be fully Plug and Play, the
BIOS, the operating system and all devices must be Plug_and_play.Windows 9x and Windows
2000/XP/2003 support Plug and Play, but Windows NT doesn’t. However, Windows 2000/XP/2003 Plug
and Play is more advanced than the Windows 9x version and doesn’t use the startup BIOS to help with
Plug and Play configurations. Therefore, on a Windows 2000/XP/2003 system, it isn’t important if the
motherboard BIOS is Plug and Play.The Plug and Play standard applies to the operating system, to BIOS
on the motherboard, and to BIOS on devices. Plug and Play under Windows or the motherboard BIOS
doesn’t assign resources to a device if the device doesn’t allow it. For example, if a legacy sound card
requires a certain group of upper memory addresses that are hard-coded into its on-board BIOS, there's
nothing that Plug and Play can do about that. (Hard-coded is computer jargon for something being coded
so that it can’t be changed.) Plug and Play simply tries to work around the problem as best it can. If two
non-Plug and Play hardware devices require the same resource, and their BIOS doesn’t provide for
accepting a substitute, these two devices can’t coexist on the same PC. Newer devices that contain BIOS
that is Plug_and_play compliant are more cooperative. At startup, they simply send a request and then
wait for the OS to assign the resources they need to work. Startup BIOS that is Plug and Play does some
of the up-front work for Windows 9x, but Windows 2000/2003 does all its Plug and Play configuration
without the help of BIOS.

Note: A Plug and Play hardware device has something like Windows 9x Ready or Windows 9x Compliant
written on the box.

ESCD (extended system configuration data) Plug and Play BIOS is an enhanced version of Plug and Play
that creates a list of all the things you've done manually to the configuration that Plug and Play doesn’t
do on its own. This ESCD list is written to the BIOS chip, so that the next time you boot, the startup BIOS
can faithfully relay that information to Windows 9x. The BIOS chip for ESCD BIOS is a special RAM chip

188
called Permanent RAM, or PRAM, that can hold data written to it without the benefit of a battery, which
the CMOS setup chip requires.
CMOS setup chip
Another chip on the motherboard, called the CMOS configuration chip, CMOS setup chip, or CMOS RAM
chip, contains a very small amount of memory, or RAM, enough to hold configuration or setup information
about the computer. This chip is responsible for remembering the current date and time, which hard
drives and floppy drives are present, how the serial and parallel ports are configured, and so forth. When
the computer is first turned on, it looks to the CMOS chip to find out what hardware it can expect to find.
The CMOS chip is powered by a trickle of electricity from a small battery located on the motherboard or
computer case, usually close to the CMOS chip itself so that, when the computer is turned off, the CMOS
chip still retains its data. The program to change CMOS setup is stored in the ROM BIOS chip and can be
accessed during the startup process. The keystrokes to enter CMOS setups are displayed somewhere on
the screen during startup in a statement such as “Press the Del key to enter setup.” Different types of
BIOS use different keystrokes. The CMOS setup doesn’t normally need to be changed, except when there’s
a problem with hardware, a new floppy drive is installed, or a powersaving feature needs to be disabled
or enabled. The CMOS setup can also hold a power-on password to help secure a system. Know that this
password isn’t the same password that can be required by a Windows OS at startup.

Exhibit 7.16: Device Manager Window

Hard drives and other secondary storage devices

A hard drive is an example of a secondary storage device, which is a device that can hold data and
instructions, but the data and instructions can only be stored there and not processed there by the CPU.
Before the instructions can be processed, they must be copied from a secondary storage device to a
primary storage device, which is memory.

In other words, you store files that contain data and programs on a hard drive, but to use these data and
programs, they must be copied into memory.

Another difference between secondary and primary storage devices is that secondary devices hold data
permanently, and primary devices hold data temporarily. What’s stored on your hard drive remains there,
even when the PC is turned off, but what’s stored in memory is lost as soon as you turn off your PC. The

189
most common secondary storage devices are hard drives, floppy disk drives, CD-ROM drives, DVD drives,
and Zip drives.

Exhibit 7.17: A PC’s hard drive subsystem

Most motherboards offer two EIDE (Enhanced Integrated Drive Electronics) connections. IDE (Integrated
Drive Electronics) is a group of standards that governs how a hard drive works, and EIDE is a group of
standards that governs how a secondary storage device, such as a hard drive, CD-ROM drive, or Zip drive,
can interface with a system. In the industry, you most often use the term IDE for both IDE and EIDE. The
connections for the two IDE channels on the motherboard are called the primary and secondary channels.
Each cable or channel can accommodate two IDE devices, such as a hard drive and a CD-ROM drive, for a
total of four IDE devices in one system. Power to the hard drive comes through a power cable from the
PC's power supply.

Directory structures and file names

The operating system is responsible for storing the files and folders on a secondary storage device, such
as a hard drive, using organizational methods called a file system.

Windows uses several different file systems.

Regardless of the file system used, every OS manages a hard drive by using directories (Windows calls
these folders), subdirectories, and files. A directory table is a list of files and subdirectories. On Microsoft
operating systems, when a hard drive is first installed and formatted, there’s a single directory table on
the drive called the root directory. For a logical drive C, the root directory is written as C:\. Note: A physical
hard drive can be divided into logical partitions, sometimes called volumes.

This root directory can hold files or other directories, which can have names such as C:\Tools. These
directories, called subdirectories, child directories, or folders, can, in turn, have other directories listed in
them. Any directory can have files and/or other directories listed in it, for example, C:\wp\data\myfile.txt.
The C: identifies the drive. If the directory had been on a floppy disk, it would have been either A: or B:.
When you write the drive and directories pointing to the location of the file, as in this example, the drive
and directories are called the path to the file. The first part of the file before the period is called the file
name (myfile), and the part after the period is called the file extension (txt), which, for Windows and DOS,

190
always has three characters or fewer. The file extension identifies the type of file, such as .doc for
Microsoft Word document files or .xls for Microsoft Excel spreadsheet files.

Exhibit 7.18: Root directory can hold files or other directories

File naming conventions

Because Windows 9x is built on a DOS core, you need to understand how to name files under DOS. Under
DOS, a file name can contain up to eight characters and the file extension can contain up to three
characters, which are separated by a period, like this: filename.ext. Characters used can be the letters a
through z, the numbers 0 through 9, or the following characters: _, ^, $, ~, !, #, %, &, - {, } , (, ), @, ', `. Be
sure not to use a space, period, asterisk (*), question mark (?), or backslash (\) in a file name or file
extension. Acceptable file extensions for program files are .com, .sys, .bat, and .exe. A program file
contains a list of instructions for the OS to follow. For example, the DOS utility program to display
information about the system is Msd.exe.

Under Windows 95 and later Windows operating systems, file names can be as long as 255 characters and
can contain spaces. Before Windows 95, only the Macintosh OS and UNIX used these long file names.
When using long file names in Windows 9x, remember that the DOS portion of the system can understand
only eight-character file names with three-character extensions. When the DOS part of the system is
operating, it truncates long file names and assigns new eight-character ones. Windows NT/2000/XP/2003
doesn’t have a DOS core and therefore does a better job of managing long file names.

File organization
By creating various directories on a hard drive, you can organize your program files and data files by
placing programs in one directory and files created by those programs in a second directory. This
organization is comparable to keeping paper records in separate folders. You can also organize files on
other secondary storage media, such as floppy disks or Zip drives.

Note: When representing directories, Windows uses a backslash, as in this example:

C:\Tools. However, UNIX uses a forward slash to represent directories. The root directory in UNIX is
written / and subdirectories are written as /etc or /bin. Also, Windows doesn’t care if directory names

191
and file names are uppercase or lowercase, but UNIX is case-sensitive: /Bin is a different directory from
/bin.

The FAT file system

The FAT (file allocation table) file system, is based on a table that’s a very long list of all the entries on a
hard drive that can hold data. Two FAT file systems used by Windows for hard drives are FAT16 and FAT32.
The 16 or 32 stand for the size of each number in the table – 16 bits or 32 bits – although, for FAT32, only
28 of them are used to hold a number. There’s one entry in the FAT for each cluster on the hard drive.

A cluster is the smallest unit of space on the hard drive used to hold data that’s addressed by the OS. An
OS puts files on the hard drive in increments of whole clusters.

All clusters on a hard drive are assigned a single number, the first useable cluster number is two. The OS
identifies a cluster by this number and knows how many bytes or characters of data a cluster can hold. All
clusters on a logical drive hold the same number of bytes.

The OS relates to the hard drive by using two tables, the FAT and the directory table. A logical drive, such
as drive C, has two copies of FAT but can have several directory tables, one for each subdirectory on the
drive. The directory table tells the OS the name of a file and the first cluster number for that file. All the
other cluster numbers for the file are kept in the FAT. For example, let’s say a file named Mydata.txt begins
at cluster 5 and requires three clusters to hold the file. The OS reads the file using the following steps.
1. The OS goes to the directory and reads the name of the file (Mydata.txt) and the first cluster
number (5).
2. The OS retrieves the contents of cluster 5 on the hard drive, which is the first segment of the file.
3. The OS turns to the FAT, looks at the fifth position in the FAT, and reads 6, which says that the next
segment of the file is in cluster 6.
4. It retrieves the second segment of the file from cluster 6 on the hard drive.
5. The OS then turns to the sixth position in the FAT and reads 10, which says the next segment of
the file is in cluster 10.
6. It retrieves the third segment of the file from cluster 10 on the hard drive.
7. The OS turns to the 10th position in the FAT and reads all 1s in the FAT entry, which says that this
is the last cluster in the file. (If the FAT is FAT16, then an entry of 16 ones is written in the FAT. If
it’s FAT32, then an entry of 28 ones is written.)

There are other organizational tables and entries that an OS uses to manage a hard drive besides the FAT
and directory tables.

192
 Exhibit 7.19: How an OS reads a file from the hard drive

3rd party system utilities

You can also use 3rd party system utilities to gather and report system information.

One of these programs is SANDRA, which provides a quick, comprehensive review of your entire PC
system, as shown in Exhibit 8-24. SANDRA also includes benchmarking and reporting tools. Next, you can
use SANDRA to examine your system.

Exhibit 7.20: SiSoft Sandra main Window

193
7.2 SOFTWARE MANAGEMENT

Next, you want to learn how the OS interfaces with other software, including BIOS, device drivers, and
applications. Because the CPU operates in two modes, real mode and protected mode, the OS boot loader
must be able to operate in each mode. This dual modality also affects the mode that an application uses.
Therefore, we begin our discussion of software by looking more carefully at each of these modes.

Real and protected operating modes

Remember that the two modes a CPU can operate in are real mode and protected mode.

There are several differences between real mode and protected mode, but the fundamental difference is
that, in real mode, there’s no memory/instruction protection from errant applications. In protected mode,
such protection is available. Privileged instructions may not be executed by ordinary applications. In real
mode, a CPU assumes that only one application or program is running at a time. This is called single
tasking, so it gives that program direct access to all hardware devices including memory. It uses a 16-bit
data path and 1 MB of memory addresses, unless a memory extender is used. A memory extender is an
OS utility program that provides an OS with memory addresses above 1 MB, called extended memory.

Exhibit 7.21: Real mode or MS mode provides single tasking

Note: Real mode is 8086 compatibility mode. Therefore, it’s limited to 16-bit data processing. Originally
the data path was only 8 bits wide. However, the processor could process 16-bits at a time, and two
fetches were needed from the 8 bit data path. In protected mode, more than one program can run at the
same time, which is a type of multitasking. In protected mode, each program can be safely contained
within its own range of resources. Here lies the meaning behind the two terms, real and protected. Real
mode means that the software has "real” access to the hardware and protected mode means that more
than one program can be running and each one is "protected” from the other(s).

In protected mode, more than one program can run, and the programs have access to memory addresses
of 4096 MB or 4 GB, depending on the motherboard, CPU, and OS being used, or sometimes more. In
protected mode, the OS doesn’t allow a program direct access to RAM but works as the mediator between
memory and programs. This allows the OS some latitude in how it uses RAM. If the OS is low on RAM, it
can store some data on the hard drive. This method of using the hard drive as though it were RAM is called
virtual memory, and data stored in virtual memory is stored in a file on the hard drive called a swap file
or page file. The OS manages the entire process, and the applications know nothing about this substitution
of hardware resources for RAM. The programs running in protected mode see only memory addresses
and have no idea where they’re located.

194
 Exhibit 7.22: Protected mode is a multitasking program

Even after protected mode became available, hardware and software needed to be backwardcompatible,
that is, able to support older technology. So real mode is still supported by today's CPUs and operating
systems. In fact, the CPU starts in real mode and must be told to switch to protected mode. For this reason,
an OS starts in real mode and commands the CPU to switch to protected mode before allowing user
interaction or loading an application. DOS and the MS-DOS mode of Windows 9x operate in real mode.
Windows 9x and Windows NT/2000/XP/2003 start out in real mode and then switch to protected mode.
The following table summarizes the differences between real mode and protected mode.

Real mode
• 16-bit data path
• Only one program runs at a time
• The CPU can access up to 1 MB of memory unless a memory extender is used
• Programs have direct access to hardware

Protected mode
• 16 or 32-bit data path
• Several programs can be loaded and running
• The CPU can access more than 1 MB to 4 GB or more of memory
• Programs access hardware by way of the operating system

How an OS uses real and protected modes

The OS must be in sync with the CPU, using the same mode it’s using. Not only must the OS be in sync
with the CPU, but applications must be compiled to run in either real or protected mode. In addition,
there’s a mode that’s a hybrid of real and protected mode that was used by older software written for
Windows 3.x. In this hybrid mode, the mode is real but the intent is to run more than one program in a
pseudo-protected environment. This unique situation that occurred with Windows 3.x is important,
because it helps to make clear the different ways in which an OS can manage applications.

195
Windows 3.x wasn’t really an OS, but neither did it at like normal application software. It used DOS as its
operating system and, therefore, is considered to be a DOS application. Windows 3.x provided an
operating environment, which refers to the overall support that it provided to application software, and
applications were installed under Windows 3.x. The two primary things that Windows 3.x provided that
DOS didn’t were a graphical user interface (GUI) and a limited form of multitasking.

16-bit and 32-bit software

Applications and device drivers written for Windows 3.x are called 16-bit Windows software. Data access
is 16 bits at a time, and each program is written so that it should not infringe on the resources of other
applications that might be running. Software applications written for Windows 95 and higher are called
32-bit drivers or 32-bit applications. In short, there are four general types of software that run on PCs:
• 16-bit DOS software is designed to run in real mode as the only program running and expecting
direct access to hardware. Under DOS, an application might attempt to serve as its own device
driver. These programs could use only the first 1 MB of memory.
• 32-bit DOS uses a 32-bit DOS extended mode program, which could run using the full available
range of memory.
• 16-bit Windows software is designed for Windows 3.x to run where other programs might also be
running. These applications might or might not attempt to access hardware resources directly and
could most likely use extended memory.
• 32-bit Windows software is designed to run in protected mode with other software and can be
loaded in extended memory. These applications never attempt to access hardware directly.

Nearly all applications and device drivers written today are 32-bit, although 16-bit software still exists, and
you must know how to support it in a Windows environment. DOS, with or without Windows 3.x, is a real
mode OS. Windows 9x is a hybrid operating system. Some of Windows 9x uses 16-bit data access (called
16-bit programs) and some uses 32-bit data access (called 32-bit programs). The Windows 9x 16bit
components exist primarily for backward-compatibility with older hardware and software. Because the
32-bit programs access twice as much data at one time as 16-bit programs do, the 32-bit programs are
faster. This fact largely explains why Windows 9x is faster than DOS. Windows NT/2000/XP/2003 is true
32-bit operating systems; all OS programs are written using 32-bit coding methods.

A 32-bit OS allows a 16-bit program to run by providing it with an environment that appears to the
program to be a 16-bit environment. This technique is called virtual real mode.

After Windows 9x is loaded, you can switch to real mode by using the Shut Down menu. Click Start, Shut
Down, and then select Restart in MS-DOS mode from the menu. You get a C prompt and a full screen from
which you can run DOS commands or applications. Windows 2000, Windows XP, and Me don’t support
accessing real mode while the OS is loaded, although recovery procedures using these operating systems
sometimes are done in real mode.

How an OS uses BIOS and device drivers

Looking back at Exhibit 8-2, you can see that an operating system uses programs designed to interact with
specific hardware devices. There are two kinds of programs used for this purpose, BIOS and device drivers.
In Exhibit 8-2, you can see that some device drivers belong to the OS, and others don’t. You can also see
that the total BIOS in a system can be located in several places in the system. The OS communicates with
simple devices, such as floppy drives or keyboards, through BIOS, and with more complex devices, such
as digital cameras or CD-ROM drives, through device drivers and BIOS. We first want to look at the BIOS
and then turn our attention to device drivers.
196
How an OS uses system BIOS
System BIOS contains the programming instructions to run the simple hardware devices that are common
to every system, specifically the keyboard, floppy disk drive, hard disk drive, and monitor. In addition,
system BIOS can be used to access the hard drive. In the case of any hardware device, an OS has a choice
of using system BIOS or device drivers. Most often it uses device drivers, because they’re faster. One
reason device drivers are faster than system BIOS is that device drivers are executed from RAM, and BIOS
is stored in ROM. RAM access is faster than ROM access. However, sometimes the OS virtualizes the hard
drive and makes it available when older software requires it.

Note: There’s a good way to determine whether the BIOS or a device driver is controlling a device. If the
device is configured using CMOS setup, most likely system BIOS controls it. If the device is configured using
the OS, most likely a driver controls it.

For example, the setup main menu for an Award BIOS system. It shows the ability to configure, or set, the
system date and time, the Supervisor Password (power-on password), floppy drive diskettes, the hard
drive, and the keyboard.

Exhibit 7.23: The Setup Utility menu for an Award BIOS system

Another setup window for BIOS that can configure serial ports, an infrared port, and a parallel port. All
these devices can be controlled by system BIOS.

On the other hand, there’s no setup window in this BIOS to control the DVD drive or Zip drive that are
installed on this system. The BIOS isn’t aware of these devices.
Instead, device drivers control these devices.

197
 Exhibit 7.24: Award BIOS setup window that configure I/O devices including ports

Note: CMOS setup windows are accessed during startup. A system displays a message at the bottom of
the screen saying something like, “Press Del to enter setup.” Pressing the indicated key launches a
program stored on the ROM BIOS microchip to change the contents of CMOS RAM.

How device drivers control hardware

Device drivers, which are software designed to interface with specific hardware devices, serve the same
functions as BIOS programs. However, they’re stored on the hard drive rather than on ROM chips, as BIOS
is. The OS provides some device drivers, and the manufacturer of the specific hardware device with which
they’re designed to interface provides other device drivers. In either case, unlike BIOS, device drivers are
usually written for a particular OS and must be rewritten to be used with another.

When you purchase a printer, DVD drive, Zip drive, digital camera, scanner, or other hardware device,
bundled with the device is a set of floppy disks or CDs that contain the device drivers, as shown in Exhibit
8-30. You must install these device drivers under the operating system so it has the necessary software to
control the device. In most cases, you install the device and then install the device drivers. There are a few
exceptions, such as a digital camera using a serial port to download pictures. In this case, you install the
software to drive the digital camera before you plug in the camera. See the device documentation to learn
what to do first.

Note: You can find device drivers in a number of sources. Some come with and are part of the operating
system, some come with hardware devices when they’re purchased, and some are provided for download
over the Internet from a device manufacturer's Web site.
There are two kinds of device drivers: 16-bit real-mode drivers and 32-bit protected mode drivers.
Windows 9x supports both, but Windows NT/2000/XP/2003 use only 32- bit drivers.

198
Device drivers under Windows 9x
Windows 9x comes with 32-bit drivers for hundreds of hardware devices. Windows automatically loads
these drivers into extended memory (memory above 1024K) at startup or when the device first needs
them. However, Windows drivers don’t support all older devices, so a system might sometimes need to
use an older 16-bit real-mode device driver. These 16-bit drivers are loaded by entries in the Config.sys,
Autoexec.bat, and System.ini files, text files used to configure DOS and Windows 3.x that are supported
by Windows 9x for backward-compatibility. These drivers use upper memory addresses. When the driver
is installed, the driver installation program makes appropriate entries in these files.

Note: Under DOS, when any program, such as a device driver, stays in memory until the CPU needs it is
called a terminate-and-stay-resident (TSR) program. The term is seldom used today, except when talking
about real mode programs.

Windows uses Autoexec.bat and Config.sys, in order to be backward-compatible with DOS, and uses
System.ini, in order to be backward-compatible with Windows 3x.

However, using 16-bit drivers can slow performance, so to get the most out of Windows 9x, use 32-bit
protected-mode drivers designed for Windows 9x. When selecting a driver to install, be sure the driver
claims to be Windows 9x compatible, which means that it’s a 32-bit driver and you can install the driver
using the Add New Hardware Wizard. If the driver must be installed using a setup program provided by
the driver manufacturer and can’t be installed using the Add New Hardware Wizard, then it’s a 16-bit
driver.

Windows 9x keeps information about 32-bit drivers in the Windows registry, a database of hardware and
software settings, Windows configuration settings, user parameters, and application settings. Windows
32-bit drivers are sometimes called dynamic drivers, because they can be loaded into memory when the
device is accessed and then unloaded to conserve memory when the device is disconnected or turned off.
Drivers that always remain in memory are called static drivers.

Sometimes, to address bugs, make improvements, or add features, manufacturers release device drivers
that are more recent than those included with Windows. Whenever possible, it’s best to use the latest
driver available for a device provided by the device manufacturer. You can usually download these
updated drivers from the manufacturer's Web site. For example, let’s say you just borrowed a printer
from a friend but you forgot to borrow the CD with the printer drivers on it. You can go to the printer
manufacturer's Web site, download the drivers to a folder on your PC, and install the driver under
Windows Sixteen-bit drivers under Windows 9x can cause slow performance, so use 32-bit drivers
whenever possible. Sometimes, it can be difficult to determine whether a device driver is 16-bit or 32bit.
One way to identify whether Windows 9x is using a 16-bit driver is to go into Device Manager and look for
an exclamation point beside the device.

This indicates that the driver has a problem. This might indicate that the driver is a 16- bit driver. If a driver
is loaded from Config.sys or Autoexec.bat, it’s a 16-bit driver; if it’s loaded from the registry, it’s a 32-bit
driver. System.ini can contain both types.

199
Device drivers under Windows 2000
Windows 2000 installs protected-mode drivers in the same way as Windows 9x, but Windows 2000
doesn’t claim to support 16-bit device drivers. For Windows 2000, always check the hardware
compatibility list (HCL) to determine if a driver works under Windows 2000. Go to the following Microsoft
Web site and search for your device:
www.microsoft.com/windows2000/professional/howtobuy/upgrading/compat
If the device doesn’t install properly or produces errors, check the manufacturer's Web site for a driver
that the manufacturer says is compatible with Windows 2000.

7.3 APPLICATION MANAGEMENT

Most application software fits into the following eight categories:

• Word processing
• Spreadsheets
• Database management
• Graphics
• Communications
• Games
• Mathematical modeling
• Software development

Each software category contains many different products. For example, some popular database
management packages include Access, Paradox, and Filemaker, and two popular word-processing
packages include Word and WordPerfect. Some application software manufacturers produce suites of
software, which combine a word-processing program and spreadsheet program and usually include a
database management program, a presentation package, an e-mail package, and a World Wide Web
browser package. Suites have many advantages, including providing similar user interfaces. The programs
are designed to make it easy to move data from one suite program to another, and files within a suite's
programs can be linked, so that updates to data or text are automatically recorded in all linked files.

Application software is designed to work on a particular OS. An application depends on an OS to provide

access to hardware resources, manage its data in memory and in secondary storage, and perform many
other background tasks. For example, consider an application running on Windows 98. Windows 98 loads
the application and executes it. The application can’t run or load itself without Windows 98. Windows 98
stays available to the application for the entire time the application is running. The application passes
certain functions to Windows 98, such as reading from a CD-ROM or printing. An application written to
work with one OS, such as Windows 98, doesn’t necessarily work with another, such as a Macintosh
system. There are, however, some exceptions. For instance, OS/2 is written so that any application
designed to work with DOS also works with OS/2, an early selling point for OS/2. However, to take full
advantage of an operating system's power and an application's power, buy application software written
specifically for your OS. Application software is available on floppy disks and CD-ROMs, or it can be
downloaded from the Internet. The application usually must be installed on a hard drive in order to run.
During the installation, the install program creates folders on the hard drive and copies files to them. For
Windows, it also makes entries in the Windows registry and puts files in the folder and subfolders that it
later uses as software components to perform various tasks. It can also place icons on the desktop and
add entries to the Start menu. Because the install program does all the work for you, installing a software
package usually is very easy.

200
Loading and initializing application software
Opening an application is a simple task for the user. For example, in Windows, you click the shortcut icon
on the desktop, and the application window opens ready for use. But what happens behind the scenes
between your click and the appearance of the application? Software or programs are stored in program
files on a secondary storage device, such as a hard drive or CD, until needed. Before they can be executed,
they must first be copied or loaded into RAM and assigned memory addresses.

Exhibit 7.25: RAM is a hardware resource; memory addresses are system resources

Even the operating system programs must be copied from the hard drive into memory before these
programs are executed.
1. The OS program files are copied into memory and assigned memory addresses and then executed.
2. The OS copies a program file into memory, which is executed.
3. The program requests data from the hard drive, which is copied into memory, and then the
program uses that data.

In addition, BIOS and device drivers are also using memory and are executed from memory as the OS
needs them.

Exhibit 7.26: Memory is used by the OS, the applications, and data

201
Understanding the process of loading and initializing software is important to people responsible for
supporting PCs. Listed below are the major steps that must take place.
1. The OS receives the command to execute the application.
2. The OS locates the program file for the application.
3. The OS loads the program file into memory.
4. The OS gives control to the program.
5. The program requests memory addresses from the OS for its data.
6. The program initializes itself and possibly requests that data from secondary storage be loaded
into memory.
7. The program turns to the user for its first instruction.

Begin with Windows 98 in MS-DOS mode as our OS. Before we discuss the specific command to load
software in real mode, let's first look at the command prompt the user sees and the information it
provides. If you boot a Windows 98 PC from a floppy disk into command prompt mode, the MS-DOS
prompt looks like this:
A:\>

This prompt is called the A prompt. The MS-DOS prompt (A:\>) displayed immediately after booting means
that the OS was copied from drive A, which is where the startup disk containing enough of the OS to load
in real mode is stored. Drive A then becomes the default drive and default directory, sometimes called
the current working drive and directory, which the OS automatically uses to save and retrieve files. The
colon following the letter identifies the letter as the name of a drive, and the backslash identifies the
directory on the drive as the root or main directory. The > symbol is the prompt symbol that the OS uses
to say, “Enter your command here.” To make the hard drive (drive C) the default drive, enter C: at the A
prompt.

The prompt now looks like this:

C:\>

Recall that when an OS first formats a hard drive for use, the format procedure creates a single directory
on the drive, called the root directory. This directory is written in the OS command lines as a single
backslash (\), or forward slash in the case of UNIX, with no other directory name following. In the
preceding OS prompt, the backslash indicates that it’s the root directory.

Starting a program
At the command prompt, when you type a single group of letters with no spaces, the OS assumes that you
want to execute a program that has the file name that you just typed and is stored in a program file in the
current directory. The OS first attempts to find the program file by that name, then copy the file into RAM,
and then execute the program. Let's use the program, Mem.exe, a Windows 9x utility to display how
memory is currently allocated, as our example. The program file, Mem.exe, is stored on the hard drive in
C:\Windows\command folder. When you type mem at the A: prompt, like this: A:\>mem

202
 Exhibit 7.27: Finding a program file

The OS says it can’t find the program to execute. It looked only on the floppy disk (drive A) for
Mem.com, Mem.exe, or Mem.bat, the three file extensions that MS-DOS recognizes for programs. If the
OS doesn't find any of these files in the current directory, it stops looking and displays the error message:
Bad command or file name

To help the OS locate the program file, you must first change the default drive to the hard drive by giving
the command:
A:\> C:

Notice, that the prompt changes to C:\>, indicating that the logical drive C on the hard drive is the default
drive. Now you change the default directory on the hard drive to \Windows\Command using the CD
(change directory) command like this:
C:\>CD\windows\command

Remember that DOS and Windows commands aren’t case-sensitive, so it makes no difference whether
you type CD, Cd, or cd. The prompt now looks like this:
C:\WINDOWS\COMMAND>

Next, we enter the mem command again, and this time the OS locates and executes the program file.
There’s another way to tell the OS where to look to find program files. As part of the boot procedure, you
can give the OS a list of paths in which to look for executable program files beyond the default directory,
by using the Path command. You can cause the Path command to be executed automatically during the
boot process by storing the command in the Autoexec.bat file. You can also execute the Path command
at any time after booting. The last Path command you execute overrides any previous ones. To see the
203
list of paths that are presently active, type Path at the command prompt, and then press Enter. To enter
a new list of paths, type Path followed by each path name, separating one path from the next by a
semicolon.

Exhibit 7.28: The path command

The first Path command displays the list of active paths, which are three logical drives, F, A, and G. The
second Path command changes this list; giving the OS the same three logical drives as well as the drive C
root directory and the C:\Windows\Command directory. The last Path command displays the new list of
active paths.

When you tell the OS to execute a program, you can also include the path to that program file as part of
the command line. For example, if the Mem.exe file is stored in the directory C:\Windows\Command, you
can execute the program by typing the following:
A:\> C:\Windows\Command\Mem.exe

Here, you’re telling the OS that the name of the program file is Mem.exe and that its location is in the
directory \Windows\Command on drive C. With this method, the directory and drive that contains the
Mem program need not be the defaults, nor do you need to use the Path command.

In summary, using the MS-DOS mode of Windows 9x, the OS searches for executable program files using
the following rules:
1. If no path is given before the file name, the OS looks in the current directory.
2. If there’s a path given in front of the file name in the command line, the OS looks in that path.
3. If no path is given, and the file isn’t in the current directory, the OS looks in the paths given to it
by the last Path command executed.

If you’ve an Autoexec.bat file in your root directory when Windows 9x starts, it reads the Path command
stored in that file. You can also store DOS commands in batch files and execute them from Windows 9x
by double-clicking the file name of the batch file in Explorer.

If you don't have an Autoexec.bat file with a Path command, Windows 9x uses a default path: C:\Windows;
C:\Windows\Command

Copying the program into memory

Recall that once the OS finds the program file, it copies the file into memory (RAM) in a location that the
OS chooses. After it copies the program into memory, the OS goes to the first address in memory occupied
by the program to receive its first instruction. If the program requests some memory for its data, and most
will, the OS decides which memory addresses to give the program, usually the memory after the program.

204
 Exhibit 7.29: Applications software is stored in files but executed from memory

Starting an application in Windows

When using the Windows GUI, once an application is installed, there are four ways to execute software:
• Double-click a shortcut icon on the desktop. You can place a shortcut icon directly on the desktop
for the applications you use often and want to get too quickly. These shortcuts contain the command
line used to execute the application. To view this command line, right-click an application icon. From
the menu that appears, select Properties. The icon's Properties box appears, from this box, you can
view the complete command line that the icon represents.
• Click the Start button, choose Programs, and then choose the program from the list of installed
software.
• Use the Run command: Click the Start button, and then choose Run to display the Run dialog box. In
this box, enter a command line or click Browse to search for a program file to execute.
• Execute a program or launch an application file by double-clicking the file name in Windows Explorer
or My Computer.

205
 Exhibit 7.30: The icon's Properties box

Exhibit 7.31: The Windows Run dialog box allows entering DOS-like commands

Installing Windows components

1. Open Control Panel -Double-click Add/Remove Programs
2. 2 Click the Windows Setup tab Verify that there are several categories of Windows components
listed in the Components field
3. Click Accessories- Click Details
4. Check Desktop Wallpaper Click OK To close the window.
5. Click Apply
6. Click OK when prompted. Type the path to the Windows setup files Click OK
7. After the installation of the files containing the new desktop wallpaper is complete, click OK To
exit the Add/Remove Programs utility.
8. Double-click Display
9. Click the Background tab Browse through the various wallpapers Experiment with the Tile, Center,
and Stretch options in the dropdown menu Observe the preview of your wallpaper selections in
the Background tab
10. Select a background combination you like Click Apply Click OK To close the Display utility.
11. Close any other open windows Observe your new desktop wallpaper 12. What collection of files
is needed to install a new Windows component?
13. Which utility is used to install new Windows components?

206
 14. What category of components is Desktop Wallpaper part of? 15.
What utility did you use to select wallpaper for your system?
16. Is it necessary to apply the Display properties before you see what wallpaper looks like?
Review Questions

1. The device in a computer that ultimately connects all hardware devices to the CPU is known as the
_____________.
2. Name three types of memory modules.
3. Which one of the following buses is designed to accommodate a fast video card? a) ISA
b) PCI
c) AGP
d) FireWire
4. How many devices can be directly connected using FireWire?
5. What’s the primary tool you use in Windows 98/2000/XP to manage buses?
6. If the command prompt is A:\> and you want the default drive to be drive C, what command do
you use?
7. What’s the command to change the default directory to \Windows?
8. If you enter the command MEM at the command prompt, what file names and extensions does
the OS look for to execute a program?

207
ASSIGNMENT.GUIDELINES.....ATTENTION STUDENTS/STAFF

If you COPY AND OR REPRODUCE SOMEBODY ELSE’S WORK WITHOUT REFERENCING, your
assignment will be penalized by 30%.
TO AVOID BEING PENALISED IN YOUR ASSIGNMENTS DO NOT

 Copy and paste information from the internet and on-line media, such as
encyclopedias or journal articles without acknowledging the source of
information (Referencing).

 Transcribe information from any textbook, encyclopedia, newspaper, journal

articles, without acknowledging the source of information (Referencing)

 Modify information without acknowledging the source of information

(Referencing)

 Use photographs, videos, or audio without permission or acknowledgement

 Get other people to write your assignment

 Buy assignments and submit them as your own effort

 Use previous work for a new assignment without citing the original assignment
in your reference list.

Types of Student Plagiarism

Academic Integrity
Good academic practice is about adopting strategies and behaviour that allow students
to complete their studies independently and honestly, and writing assignments in an
appropriate academic style. The task submitted is also awarded better marks! Bad academic
practice includes dishonesty, cheating and plagiarism and also work that is badly prepared and
rushed.
Direct copying
This sort of plagiarism is taking the exact words somebody else wrote, (in a website, a book,
another student's work, or any other source) and putting that into an assignment, without
referencing that somebody else’s work has been cited.
Word-switching
This sort of plagiarism means taking someone else's writing and changing words here and there,
or taking little bits of sentences, without acknowledging whose ideas have been adopted.
Concealing sources
Not making it obvious when drawing on somebody else's work will be regarded as plagiarism.
This includes:
 Taking somebody else's ideas and putting them into your words without telling us
where you got the ideas.
 Using a source several times, but only referencing it once. Using ideas from the same
source several times in a piece of work, requires a citation each time the source is used.
Working with other students
It is preferable for students to do their assignments themselves. Hence
 Copying another student's work is plagiarism.
 Submitting all or part of another student's work as your own is plagiarism.
 Sharing written work is plagiarism.
 Paying somebody to do your work for you is plagiarism.
 In an individual assignment, writing the assignment with other people is plagiarism.
(Group assignments are different!).
 Asking another student to translate your ideas into English, or getting their help to
write your assignment is plagiarism.
Unless you are told to work in a group, you must work alone. If you want to talk to your friends
about the work, do it before you start writing. The work you submit must be your own!
When any of the above types are identified, each case will be investigated for:

 Intention of student
 Degree of plagiarism – how much of the assignment has been plagiarized
 Previous offences by the student