Department of Computer Science and Engineering

CLOUD COMPUTING (BCS601)

Module-03

Cloud Platform Architecture over Virtualized Datacenters

4.1 Cloud Computing and Service Models

 Over the past 20 years, the global economy has shifted from manufacturing to service

.IN
industries.
 By 2010, 80% of the U.S. economy was service-based, with only 15% in manufacturing
and 5% in agriculture and other sectors.
C
 Cloud computing is especially beneficial for the service industry and brings a new way of
N
doing business computing.
SY

 In 2009, the global cloud market was worth $17.4 billion; it was predicted to grow to
$44.2 billion by 2013.
 Cloud application developers rent resources from large automated data centers instead
U

of buying expensive hardware.

 Cloud platforms allow users to access and use applications from anywhere at low costs.
VT

 These platforms are usually built on top of large data centers using virtualization.
 Cloud computing turns data centers into virtual systems with automated management
of hardware, databases, interfaces, and applications.
 The goal of cloud computing is to improve data centers using automation and efficient
resource use.

4.1.1 Public, Private, and Hybrid Clouds

 Cloud computing came from older systems like cluster, grid, and utility computing.
 Cluster and grid computing use many computers working together.
 Utility computing and SaaS let you pay only for what you use.
 Cloud computing gives services using big data centers.
 People can use cloud services from anywhere, anytime.
Dept. of CSE, SVIT Page 1
  Instead of moving data around, cloud sends programs to the data.
 This saves time and improves internet speed.
 Virtualization helps use resources better and cuts costs.
 Companies don’t need to set up or manage servers themselves.
 Cloud provides hardware, software, and data only when needed.
 The goal is to replace desktop computing with online services.
 Cloud can run many different apps at the same time easily.

.IN
4.1.1.1 Centralized versus Distributed Computing

 Cloud computing is distributed using virtual machines in big data centers.

C
 Public and private clouds work over the Internet.
 Big companies like Amazon, Google, and Microsoft build distributed cloud systems for
N
speed, reliability, and legal reasons.
 Private clouds (within companies) can connect to public clouds to get more resources.
SY

 People may worry about using clouds in other countries unless strong agreements (SLAs)
are made.

4.1.1.2 Public Clouds

U

 A public cloud is available to anyone who pays for it.

VT

 It is run by cloud providers (like Google, Amazon, Microsoft, IBM, Salesforce).

 Users subscribe to use services like storage or computing power.
 Public clouds let users create and manage virtual machines online.
 Services are charged based on usage (pay-as-you-go).

4.1.1.3 Private Clouds

 A private cloud is built and used within one organization (not public).
 It is owned and managed by the company itself.
 Only the organization and its partners can access it — not the general public.
 It does not sell services over the Internet like public clouds do.
 Private clouds give flexible, secure, and customized services to internal users.
 They allow the company to keep more control over data and systems.
 Private clouds may affect cloud standard rules, but offer better customization for the
company.

Dept. of CSE, SVIT Page 2

4.1.1.4 Hybrid Clouds

 A hybrid cloud combines both public and private clouds.

 It allows a company to use its private cloud but also get extra power from a public
cloud when needed.
 Example: IBM’s RC2 connects private cloud systems across different countries.
 Hybrid clouds give access to the company, partners, and some third parties.
 Public clouds offer flexibility, low cost, and standard services.
 Private clouds give more security, control, and customization.
 Hybrid clouds balance the two, making compromises between sharing and privacy.

.IN
C
N
SY
U
VT

4.1.1.5 Data-Center Networking Structure

 The core of a cloud is a server cluster made of many virtual machines (VMs).
 Compute nodes do the work; control nodes manage and monitor cloud tasks.
 Gateway nodes connect users to the cloud and handle security.
 Clouds create virtual clusters for users and assign jobs to them.
 Unlike old systems, clouds handle changing workloads by adding or removing resources
as needed.
 Private clouds can support this flexibility if well designed.

Dept. of CSE, SVIT Page 3

 Data centers use thousands to millions of servers (e.g., Microsoft’s Chicago center has
100,000 servers).
 Supercomputers and data centers are similar but have key differences.

 Supercomputers use separate storage and custom networks (like fat trees or 3D
torus).
 Data centers use server disks, memory, and standard IP networks (like 10 Gbps
Ethernet).

 Data center networks have layers:

 Layer 2: server racks connected by fast switches

.IN
 Layer 3: access and border routers that link to the Internet.
C
N
SY
U
VT

4.1.1.6 Cloud Development Trends

 Private clouds will grow faster than public clouds in the future.
 Private clouds are more secure and trustworthy for companies.
Dept. of CSE, SVIT Page 4
  Private clouds may become public clouds or hybrids as they mature.
 Hybrid clouds (mix of private and public) will be common in the future.
 Applications (like email) use service-access nodes to connect to internal cloud
services.
 Supporting service nodes help manage cloud tasks (like locking services).
 Independent service nodes provide specific services, like geographical data.
 Clouds improve network efficiency by reducing data movement.
 Clouds help solve the petascale I/O problem (handling large data).
 Cloud performance and Quality of Service (QoS) are still being tested in real

.IN
use.
C
4.1.2 Cloud Ecosystem and Enabling Technologies
N

 Traditional computing: Buy and manage hardware every 18 months.

SY

 Cloud computing: Rent resources, only pay for what you use.
 Cloud saves 80-95% compared to traditional computing.
U

 Small businesses can avoid high upfront costs with cloud.

 Cloud market may reach $126 billion by 2012.
VT

 Clouds work as service factories with many data centers.

 Cloud computing reduces costs and barriers.

4.1.2.1 Cloud Design Objectives

 Shift from desktops to data centers: Move computing, storage, and software from local
devices to data centers over the Internet.
 Service provisioning and economics: Cloud providers offer services with SLAs,
focusing on efficiency and pay-as-you-go pricing.
 Scalability: Cloud platforms must be able to scale as the number of users grows.
 Data privacy protection: Ensure data privacy to build trust in cloud services.
 High quality of services: Standardize Quality of Service (QoS) to ensure
interoperability between different cloud providers.

Dept. of CSE, SVIT Page 5

  New standards and interfaces: Solve the data lock-in problem with universal APIs for
flexibility and portability.

4.1.2.2 Cost Model

Traditional IT costs:

 Users buy hardware and face fixed costs and operational costs (e.g., maintenance).
 Costs increase with more users.

Cloud computing costs:

.IN
 Pay-as-you-go model: No upfront cost, only pay for operational expenses.
 Cloud computing is cheaper because it avoids large initial investments and scales with
demand.
C
Cloud computing is great for small businesses:
N
 No need to buy expensive equipment, just pay for what you use.
 Ideal for businesses that need flexibility and want to avoid heavy capital expenses.
SY

Cloud Ecosystems:

 Cloud computing has created an ecosystem of providers, users, and technologies.

U

 Private and hybrid clouds are growing, offering flexible resources with public cloud
involvement.
VT

Private Cloud Ecosystem:

 Four levels: Consumer demands, cloud management, virtual infrastructure

management, and VM management.
 Ecosystem tools help manage cloud resources but lack standard interfaces.

Cloud tools:

 VM management tools like vSphere, oVirt, and OpenNebula help manage VMs and
virtualized resources.
 Public cloud tools like Amazon EC2 and Eucalyptus support cloud infrastructure.

4.1.2.4 Surge of Private Clouds

 Private clouds use a company’s existing IT setup and staff.

 Both public and private clouds handle changing workloads.
 Public clouds work best with independent tasks (no communication between jobs).
Dept. of CSE, SVIT Page 6
  Private clouds balance workloads within the company’s network for better efficiency.
 Private clouds offer better security, privacy, and testing environments.
 Public clouds help avoid big upfront costs in hardware, software, and staff.
 Companies often start by virtualizing their systems to reduce operating costs.
 Big companies (like Microsoft, Oracle, SAP) use policy-based IT management to
improve services.
 IT as a service boosts flexibility and avoids replacing servers often.
 This leads to better IT efficiency and agility for companies.

.IN
4.1.3 Infrastructure-as-a-Service (IaaS)

 IaaS means renting IT infrastructure like servers, storage, and networks over the
C
Internet.
 It provides virtual machines, storage, networks, and firewalls.
N
 Users can choose their own operating system and software.
 Users don’t manage physical hardware, only virtual resources.
SY

 It's a pay-as-you-go model – no need to buy expensive equipment.

 IaaS is scalable – add or remove resources anytime.
 Great for startups, developers, and large businesses.
Helps with testing, hosting apps, data backup, and disaster recovery.
U


 Examples of IaaS providers:
VT

o Amazon EC2, S3
o Microsoft Azure VMs
o Google Compute Engine
o IBM Cloud
o Oracle Cloud Infrastructure (OCI)

 Saves money and time by avoiding hardware setup.

 Ideal for companies needing flexible and powerful IT resources.

4.1.3 Platform-as-a-Service (PaaS)

 Paas provides a platform to build, test, and deploy applications.

 It includes tools, libraries, databases, and runtime environments.

Dept. of CSE, SVIT Page 7

  Developers don’t manage servers, storage, or infrastructure.
 Focus is only on writing and running code.
 It handles app hosting, scaling, updates, and security automatically.
 Faster development because everything is ready to use.
 Great for developers and software teams.
 Useful for web apps, mobile apps, and APIs.
 Examples of PaaS providers:

 Google App Engine

.IN
 Microsoft Azure App Service
 Heroku
 IBM Cloud Foundry
 Red Hat OpenShift
C
 Pay only for what you use – no upfront setup or hardware costs.
N
 Helps teams collaborate easily and launch apps faster.
SY

4.1.4 Software-as-a-Service (SaaS)

 SaaS means using software over the internet.

U

 No need to install or update anything.

VT

 Accessible from any device with internet.

 Software is managed by the provider, not the user.
 Users pay monthly or yearly (subscription model).
 No hardware or server needed by the user.
 Used for email, file sharing, CRM, video calls, etc.
 Data is stored in the cloud by the provider.
 Saves time, money, and effort.
 Great for businesses and individuals.
 Examples of SaaS:

 Gmail
 Google Docs
 Microsoft 365
 Salesforce
 Zoom
Dept. of CSE, SVIT Page 8
 .IN
C
N
SY

4.2 DATA-CENTER DESIGN AND INTERCONNECTION NETWORKS

A data center is a facility that houses a large number of servers connected through a high-
U

speed network to provide computing, storage, and service resources. A modular data center is
a smaller, portable version that fits inside a 40-foot container and can be easily deployed and
VT

expanded, offering flexibility and scalability for remote or rapid deployment needs.

4.2.1 Warehouse-Scale Data-Center Design

 Large Data Centers: Cloud computing uses huge data centers, sometimes as big as
shopping malls, with hundreds of thousands of servers.
 Lower Costs for Bigger Centers: Bigger data centers are cheaper to run per server due
to economies of scale.
 Cost Breakdown: Operating a large data center is cheaper, with lower network and
storage costs compared to small ones.
 Microsoft's Global Data Centers: Microsoft runs around 100 data centers worldwide to
support cloud services.

Dept. of CSE, SVIT Page 9

4.2.1.1 Data-Center Construction Requirements

 Data Center Components: Data centers use off-the-shelf components, such as servers
with multi-core CPUs, DRAM, and disk drives. Servers are connected via switches for
resource access.
 Bandwidth and Latency: Disk bandwidth varies between local and off-rack storage.
Large data centers often face challenges due to discrepancies in latency, bandwidth, and
capacity.
 Failure and Reliability: Hardware and software failures are common in large data

.IN
centers. Redundant hardware and data replication ensure reliability, preventing data
loss during failures.
 Cooling System: Data centers use raised floors to distribute cool air to server racks.
C
Cold air is pumped into the floor and escapes through perforated tiles in front of racks.
N
Hot air is then returned to the cooling units for recirculation. Advanced systems may
also use cooling towers for efficiency.
SY
U
VT

Dept. of CSE, SVIT Page 10

4.2.2 Data-Center Interconnection Networks

The interconnection network in a data center connects all servers in the cluster and is a vital component for
performance.

The network must meet five key requirements:

1. Low Latency: Ensure fast communication between servers.

2. High Bandwidth: Enable data transfer with high throughput.
3. Low Cost: Keep the network affordable for large-scale implementations.
4. MPI Support: Support efficient message-passing interface (MPI) communication for parallel processing.
5. Fault Tolerance: Maintain network functionality even in the event of failures.

.IN
The network must handle both point-to-point (direct server-to-server) and collective communication (group
communication) among servers.

4.2.1 Warehouse-Scale Data-Center Design

C
 Cloud computing runs on big buildings called data centers.
N
 Some data centers are huge—as big as shopping malls or 11 football fields.
 A large data center can hold 400,000 to 1 million servers.
SY

 Small data centers might only have 1,000 servers.

 Bigger data centers are cheaper to run per server (this is called "economies of scale").
 Running a large data center is cheaper than running a small one.
 For example, in a 400-server data center:
U

 Network cost is $13 per Mbps


VT

Storage cost is $0.40 per GB

 A small data center has:

 Network costs 7 times higher

 Storage costs 5.7 times higher

 Microsoft has about 100 data centers around the world.

4.2.1.1 Data-Center Construction Requirements

 Most data centers use standard, off-the-shelf components (not custom-built supercomputing parts).
 Each server has:

 Multiple CPUs (with multiple cores)

 8 GB of memory (DRAM)
 Four 1 TB disk drives

Dept. of CSE, SVIT Page 11

  Servers are grouped in racks (40 servers per rack) and connected through rack-level switches.
 All racks are connected to a cluster-level switch.
 Disk speed differences:

 Local disk: 200 MB/s

 Off-rack disk (via shared link): 25 MB/s

 The total disk storage is about 10 million times larger than the memory (DRAM).
 Big applications must handle differences in speed and size between memory and storage.
 Large-scale data centers are more cost-efficient, but...
 Failures are common (about 1% of servers can fail at any time):

 Hardware issues (CPU, disk, network)

.IN
 Software bugs

 Power failures can shut down the entire data center

 To avoid losing data, systems must:
C
 Use redundant hardware
N
 Store multiple copies of data in different places
 Keep data available even during failures (high reliability).
SY

4.2.1.2 Cooling System of a Data-Center Room

 Data centers use raised floors (2–4 feet above concrete) to hide cables and distribute cool air.
U

 Cold air is blown under the floor by CRAC units (Computer Room Air Conditioning).
 Cold air comes out through perforated tiles in front of server racks.
VT

 Server racks are arranged in cold and hot aisles:

o Cold aisle: where cold air enters servers
o Hot aisle: where hot air exits
 Hot air is pulled back into the CRAC units, cooled, and recycled into the floor.
 Coolant enters at 12–14°C, warms up, and then returns to a chiller.
 Newer data centers use cooling towers to help cool the system more efficiently:
o Water absorbs heat from the coolant in a heat exchanger
o The heated water is cooled in towers and reused
 In short, data centers use a carefully designed air and water system to keep servers from overheating by
constantly recycling and cooling air and liquid.

4.2.2 Data-Center Interconnection Networks

A critical part of data center design is the interconnection network that links all servers within the cluster. This
network must be carefully designed to meet five key requirements: low latency for fast communication, high
bandwidth to handle large volumes of data, low cost to maintain affordability, support for the Message Passing
Interface (MPI) used in parallel processing, and fault tolerance to ensure the system keeps running even when some
components fail. The inter-server network must efficiently handle both point-to-point communication between
Dept. of CSE, SVIT Page 12
individual servers and collective communication among multiple servers at once. Meeting these needs is essential
for the smooth and reliable operation of large-scale data centers.

4.2.2.1 Application Traffic Support

 The network must support all MPI communication types (point-to-point and collective).
 It should have high bisection bandwidth to handle large data flow across the network.
 One-to-many communication is needed for tasks like distributed file access.
 Metadata master servers must talk to many slave servers.
 The network should support MapReduce operations at high speed.
 It must handle different types of network traffic used by various applications.

4.2.2.2 Network Expandability

.IN
C
 The network must be expandable to support thousands or more servers in the future.

N
Topology should allow easy restructuring as more servers are added.
 It should support load balancing and smooth data movement to avoid bottlenecks.

SY

Fat-tree and crossbar networks can be built using low-cost Ethernet switches.
 As server numbers grow, network design becomes more complex.
 Modular growth is important—server containers are used as building blocks.
 Each data-center container can hold hundreds to thousands of servers.
U

 Containers are pre-built units—just connect power, network, and cooling to start.
 This reduces setup and maintenance costs.
VT

 Networks can be built by:

 Creating a core connection backbone first, then linking end servers.

 Using external switching and cabling to connect multiple containers.

4.2.2.3 Fault Tolerance and Graceful Degradation

 The network must handle link or switch failures without crashing.

 There should be multiple paths between any two servers for reliability.
 Redundancy (extra hardware/software) helps tolerate failures.
 Data and computing are replicated across servers for fault tolerance.
 Similar redundancy should apply to the network (hardware and software).
 Software should detect failures and reroute traffic around broken links.
 This should happen automatically and without affecting cloud services.
 The network should degrade gracefully during small failures.
Dept. of CSE, SVIT Page 13
  Hot-swappable components (easy to replace while running) are preferred.
 The design should avoid any single point of failure.
 Networks are usually built in two layers:

 Lower layer connects to end servers

 Upper layer connects server groups (backbone network)

 This hierarchical structure supports modular container-based data centers.

4.2.2.4 Switch-centric Data-Center Design

 There are two approaches to building data-center-scale networks:

.IN
1. Switch-centric: Uses switches to connect servers, no changes needed on the servers themselves.
2. Server-centric: Modifies the operating system on servers with special drivers to handle traffic.
C
 Fat-tree network design is an example used in data centers:
N
 Two layers:
SY

o Bottom layer: Server nodes connected by edge switches.

o Upper layer: Aggregates the edge switches through aggregation switches.

 Pods: Each pod contains edge switches, aggregation switches, and their connected nodes.
U

 Core switches link different pods.

 Multiple paths between any two servers provide fault tolerance if links fail.
VT

 Key benefits of the fat-tree design:

 Switch failure doesn't affect the entire network (alternate paths exist).
 Edge switch failure affects only a small number of servers.
 Higher bandwidth within pods supports massive data movement for cloud applications.
 Low-cost Ethernet switches are used, making the design more affordable.
 Routing algorithms inside switches help find alternate paths if there’s a failure.
 Server nodes aren't affected by switch failures as long as alternate paths are intact.

4.2.3 Modular Data Center in Shipping Containers

 Modern data centers are often made up of truck-towed containers that house server clusters.
 SGI ICE Cube modular data centers have containers that hold hundreds of blade servers in racks, with
fans circulating heated air through a heat exchanger to cool it for the next rack.
 A single container can hold 46,080 processing cores or 30 PB of storage.

Dept. of CSE, SVIT Page 14

  Container-based data centers offer benefits like:

 Lower power consumption

 Higher computing density
 Mobility (easy to move to areas with cheaper electricity or better cooling)

 Cooling costs can be reduced by 80% compared to traditional data centers through efficient chilled air and
cold water circulation.
 These data centers are often built in locations with cheaper utilities and more efficient cooling.
 Modular containers can form a large-scale data center, like a shipping yard of containers.
 Centralized management (in a single building) is important for handling data integrity, server

.IN
monitoring, and security management.

4.2.3.1 Container Data-Center Construction

C
 Modular data centers are housed in truck-towable containers that include all necessary components like
N
network, computer, storage, and cooling systems.
 Cooling efficiency can be improved by adjusting water and airflow with better airflow management.
SY

 Data center construction evolves in stages: starting with a single server, moving to a rack system, and
finally to a container system.
U

 Building a rack of 40 servers might take half a day, but expanding to a container system with 1,000
servers requires more time for layout, power, networking, cooling, and testing.
VT

 The container must be weatherproof and easily transportable.

 Modular construction can be completed quickly (a few days) if all components and supplies are ready.
 This design is suitable for many cloud applications, like in the healthcare industry, where data centers can
be installed at clinic sites.
 Information exchange with a central database and ensuring data consistency becomes a challenge in
hierarchical data center structures.
 Security in cloud services may require the use of multiple data centers.

Dept. of CSE, SVIT Page 15

 .IN
4.2.4 Interconnection of Modular Data Centers

 Container-based data-center modules can be combined to build large-scale data centers using multiple
container modules.
C
 One example of such a network design is the server-centric BCube network for modular data centers.
 The BCube network uses a layered structure:
N
 Level 0 consists of server nodes.
SY

 Level 1 contains switches connecting to Level 0 servers.

 The structure is recursively built, where each BCubek is made up of n BCubek-1 networks with n-port
switches.
U

 In BCube1:
VT

 Each server in BCube0 connects to the corresponding port of a Level 1 switch.

 Servers have multiple ports attached, allowing additional devices to connect.

 The BCube network provides multiple paths between any two nodes, offering:

 Extra bandwidth for different cloud application communication patterns.

 Routing support via a kernel module in the server’s OS, allowing packet forwarding without modifying
upper layer cloud applications.
 This design allows cloud applications to run on top of the BCube network without requiring changes.

4.2.4.1 Inter-Module Connection Networks

 BCube networks are used inside server containers in modular data centers.
 However, to connect multiple containers, a new level of networking is required, leading to the design of
MDCube (Modularized Datacenter Cube).

Dept. of CSE, SVIT Page 16

  MDCube connects multiple BCube containers using high-speed switches inside each BCube, creating a
network topology for inter-container communication.
 MDCube builds a virtual hypercube at the container level, complementing the internal cube structure of
each BCube.
 The MDCube network is scalable, supporting large-scale data centers and cloud applications.
 The architecture supports cloud application communication patterns and helps create a high-
performance inter-container network.
 For detailed information, readers can refer to the full study by Wu et al. [82], which includes
implementation and simulation results of this design.

4.2.5 Data-Center Management Issues

.IN
C
 Quality Service for Users: The data center should ensure quality service for users for at least 30 years.
 Controlled Information Flow: The system should streamline information flow and focus on high
N
availability and consistent services.
SY

 Multiuser Manageability: The system must handle all data center functions, such as traffic flow, database
updates, and server maintenance.
 Scalability: The system should grow with increasing workload, with scalable storage, processing, I/O,
U

power, and cooling.

VT

 Reliability in Virtualized Infrastructure: Features like failover, fault tolerance, and VM live migration
should be integrated to ensure recovery from failures or disasters.
 Low Cost: Minimize costs for both users and providers, including operational expenses.
 Security and Data Protection: Strong security measures should be in place to protect against attacks,
ensure data privacy, and maintain integrity.
 Green Technology: Data centers should focus on energy efficiency and reducing power consumption.

4.2.5.1 Marketplaces in Cloud Computing Services

 Factory Racking and Packing: Data centers should be built efficiently in factories, avoiding complex
packaging layers at the customer site.
 Custom-Crafted vs. Prefabricated: While modular designs are more space-efficient, data centers are still
typically custom-crafted rather than being prefab units.
 High Power Density: Modular data centers can support high power densities, exceeding 1250 W/sq ft.
Dept. of CSE, SVIT Page 17
  Flexible Installation Locations: These data centers can be installed on rooftops or parking lots.
 Future Upgrades: It's important to include enough redundancy and flexibility in the design to allow for
future upgrades as needs grow.

4.3 ARCHITECTURAL DESIGN OF COMPUTE AND STORAGE CLOUDS

4.3.1 A Generic Cloud Architecture Design

An Internet cloud is envisioned as a public cluster of servers provisioned on demand to perform collective web
services or distributed applications using data-center resources.

.IN
4.3.1.1 Cloud Platform Design Goals

 Scalability: The system should easily expand by adding more servers and network capacity as needed,
C
supporting growing workloads and user demands.
 Virtualization: Cloud management must support both physical and virtual machines, allowing flexible
N
resource allocation and efficient operation.
SY

 Efficiency: The platform should be optimized for performance and resource use, ensuring cost-effectiveness
and smooth operations.
 Reliability: Data should be replicated across multiple locations, so even if one data center fails, the data
U

remains accessible. This ensures system uptime and continuous service.

VT

4.3.1.2 Enabling Technologies for Clouds

Driving Forces Behind Cloud Computing:

1. Ubiquitous Networking: Widespread broadband and wireless networking allow seamless access to cloud
services.
2. Falling Storage Costs: Decreasing prices for storage make it more affordable to build and maintain large-
scale data centers.
3. Improvements in Internet Software: Advancements in software have enhanced cloud capabilities, making
services more reliable and efficient.

Benefits to Cloud Users and Providers:

Dept. of CSE, SVIT Page 18

  Users can easily scale their resources, reduce costs, experiment with new services, and discard unused
capacity.
 Service Providers can increase resource utilization through technologies like multiplexing, virtualization,
and dynamic provisioning.

Technologies Enabling Cloud Computing:

1. Hardware Advancements: Improvements in multicore CPUs, memory chips, and disk arrays allow for
faster and more powerful data centers with vast storage capacity.

.IN
2. Resource Virtualization: Virtualization enables quick cloud deployment and helps with disaster recovery.
3. Service-Oriented Architecture (SOA): Supports flexible cloud service design and integration.
4. Software as a Service (SaaS) & Web 2.0 Standards: These have enabled cloud applications and services
C
to be easily accessible and scalable.
N
5. Internet Performance: Better network infrastructure ensures fast and reliable cloud access.
6. Large-Scale Distributed Storage: A foundational technology for handling vast amounts of data across
SY

cloud environments.
7. License Management and Billing: Advances in managing licenses and automating billing help streamline
cloud service operations.
U

4.3.1.3 A Generic Cloud Architecture

VT

 Cloud Platform Structure:

 The cloud is made up of many servers that are added or removed as needed.
 These servers can either be physical machines or virtual ones (VMs).

 Key Components:

 User Interfaces: These are used by people to request cloud services.

 Provisioning Tool: It allocates the needed resources to fulfill those requests.
 Storage: Data is stored in distributed systems, like SANs.
 Database Systems: Manage the cloud’s data.
 Security Devices: Protect the cloud with things like firewalls.
 APIs: Allow developers to connect with cloud services.
 Monitoring Tools: Keep track of how resources are used and if everything is working well.

Dept. of CSE, SVIT Page 19

 Management of Resources:

 The software in the cloud automatically handles resources, like adding or removing servers.
 Big companies like Google and Microsoft have data centers around the world to make their clouds work
efficiently.

 Cloud Types:

 Private Clouds: For internal use by a single organization.

 Public Clouds: Open to everyone, managed by third-party companies.
 Hybrid Clouds: A mix of private and public clouds for special needs.

.IN
 Security Concerns:

 Trust & Reputation: Systems that ensure resources are safe and reliable.
C
 Security Monitoring: Keeps the cloud safe from attacks and breaches.
 Privacy Issues: Protecting the data and ensuring only authorized access.
N
SY
U
VT

Dept. of CSE, SVIT Page 20

4.3.2 Layered Cloud Architectural Development

cloud Architecture – 3 Main Layers

1. Infrastructure Layer (IaaS)

 Base layer of cloud architecture.

 Provides virtual servers, storage, and networking.
 Uses virtualization to manage physical hardware.
 Example: Amazon EC2.

2. Platform Layer (PaaS)

.IN
 Middle layer built on top of infrastructure.
 Offers tools for developing, testing, and running apps.
 Ensures scalability, security, and reliability.
C
 Acts like a software environment for developers.

3. Application Layer (SaaS)

N


SY

Top layer where users access software directly.

 Includes apps for office work, CRM, finance, etc.
 Example: Salesforce.com.
U

Cloud Deployment Types

 Public Cloud – via Internet (e.g., Google Cloud).

VT

 Private Cloud – via internal network.

 Hybrid Cloud – mix of both.

Dept. of CSE, SVIT Page 21

 .IN
C
N
SY

4.3.2.1 Market-Oriented Cloud Architecture

U

 Users need guaranteed Quality of Service (QoS) from cloud providers.

 QoS expectations are defined in Service Level Agreements (SLAs).
VT

 Traditional resource management is not suitable for cloud environments.

 A market-oriented approach balances resource supply and demand.
 Economic incentives help reduce costs and improve service quality.
 Users or brokers send service requests to the cloud from anywhere.
 SLA Resource Allocator manages interaction between users and data centers.
 Service Request Examiner checks QoS needs and accepts/rejects requests.
 VM Monitor tracks availability and usage of virtual machines (VMs).
 Service Request Monitor observes the progress of running tasks.
 Pricing mechanism sets charges based on time, demand, and availability.
 Accounting mechanism logs actual usage and calculates final costs.
 Dispatcher runs accepted tasks on assigned VMs.
 Multiple VMs can run on one physical machine with different OS environments.

Dept. of CSE, SVIT Page 22

  VMs can be created or stopped on demand for flexibility.
 Resource allocation is based on real-time needs and historical data.
 The system prevents overload and ensures efficient use of resources.
 This architecture supports fair pricing, scalability, and QoS delivery.

4.3.2.2 Quality of Service Factors

 Data centers contain multiple servers to meet cloud service demands.

 Cloud services support business operations and must meet QoS needs like time, cost, reliability, and
security.

.IN
 QoS requirements are not fixed and can change over time.
 Cloud systems must prioritize customer needs since customers pay for the services.
C
 Current cloud systems lack strong support for dynamic SLA negotiation.
 SLA negotiation mechanisms are needed to handle changing user demands and alternate offers.
N
 Clouds must support customer-driven service management based on individual profiles and needs.
SY

 Computational risk management helps identify and manage risks in service execution.
 Resource management should use market-based strategies to balance user needs and system efficiency.
 Autonomic resource management allows the system to adapt to changes without manual intervention.
U

 Virtual machines (VMs) enable flexible and dynamic resource allocation.

VT

 Resources are shared and adjusted in real-time based on service demands.

 The system must handle multiple competing service requests effectively.
 Overall goal: maintain SLA commitments while adapting to evolving conditions and customer expectations.

4.3.3 Virtualization Support and Disaster Recovery

One very distinguishing feature of cloud computing infrastructure is the use of system virtualization and the
modification to provisioning tools. Virtualization of servers on a shared cluster can consolidate web services. As the
VMs are the containers of cloud services, the provisioning tools will first find the corresponding physical machines
and deploy the VMs to those nodes before scheduling the service to run on the virtual nodes.
In addition, in cloud computing, virtualization also means the resources and fundamental infrastructure are
virtualized. The user will not care about the computing resources that are used for providing the services. Cloud
users do not need to know and have no way to discover physical resources that are involved while processing a
service request. Also, application developers do not care about some infrastructure issues such as scalability and
fault tolerance (i.e., they are virtualized). Application developers focus on service logic.

Dept. of CSE, SVIT Page 23

4.3.3.1 Hardware Virtualization

.IN
C
N
SY
U
VT

 Virtualization software simulates hardware to run unmodified operating systems.

 It enables legacy software and older OSes to run in modern cloud environments.
 Developers can use any OS or programming language, increasing flexibility.
 Development and deployment environments can now be identical, reducing runtime errors.
 Virtualization supports running third-party programs on cloud platforms.
 VMs offer flexible runtime environments, freeing users from hardware/software setup worries.
 Traditional resource sharing (users/groups) lacks flexibility and customization.
 VMs allow users full privileges and system customization without affecting others.

Dept. of CSE, SVIT Page 24

  Multiple VMs can run different OSes on the same physical server.
 Virtual disk storage and virtual networks are created for each VM.
 Virtualized resources are pooled and managed by special servers.
 Virtualized integration managers handle system tasks like load balancing, security, and provisioning.
 Each VM platform can deliver custom virtual solutions for different user needs.
 Virtualization ensures complete separation and isolation among users.
 Resource management includes mirror, system, user, data, and security management.
 Recovery from failures is faster with VMs compared to traditional disaster recovery methods.
 AWS uses Xen VMs; Azure uses common language runtime; Google App Engine uses Python handlers.

.IN
 Storage models vary—AWS has block and blob stores, Azure uses SQL Data Services, GAE uses BigTable.
 Network configurations are mostly automated and hidden from users, with scaling managed internally.
C
4.3.3.2 Virtualization Support in Public Clouds
N
 AWS offers full VM-level virtualization, giving users high flexibility to run custom applications.
SY

 GAE (Google App Engine) offers limited, application-level virtualization, restricting users to Google’s
predefined services.
 Microsoft Azure provides programming-level virtualization through the .NET framework.
U

 VMware tools support virtualization on workstations, servers, and data centers.

VT

 Microsoft virtualization tools are designed for PCs and some specific servers.
 XenEnterprise tools are used for Xen-based server virtualization only.
 The IT industry is widely adopting cloud computing due to its benefits.
 Virtualization supports high availability (HA), disaster recovery, dynamic load balancing, and
resource provisioning.
 Cloud computing and utility computing both rely on virtualization for scalability and automation.

4.3.3.3 Storage Virtualization for Green Data Centers

 IT power consumption in the U.S. has more than doubled, now using 3% of the nation’s total energy.
 A major cause is the high number of power-hungry data centers.
 Over half of Fortune 500 companies are adopting new energy-saving policies.
 Virtualization significantly reduces energy costs by lowering physical server usage.
 Surveys by IDC and Gartner confirm virtualization’s role in cutting power consumption.
Dept. of CSE, SVIT Page 25
  The IT industry is becoming more energy-conscious due to rising power concerns.
 There is a growing need to save energy because alternative energy options are limited.
 Virtualization and server consolidation help reduce the number of physical machines needed.
 Green data centers aim to use energy-efficient infrastructure.
 Storage virtualization adds to energy savings by optimizing storage usage.
 Together, these efforts support the goal of green computing.

4.3.3.4 Virtualization for IaaS

 VM technology allows users to build custom environments on physical infrastructure.

.IN
 Underutilized servers can be consolidated into fewer machines, saving resources.
 VMs can run legacy code without affecting other system interfaces or APIs.
C
 VMs improve security by using sandboxes to isolate risky applications.
 VMs support performance isolation, enabling better QoS guarantees for customers.
N
SY

4.3.3.5 VM Cloning for Disaster Recovery

 Traditional recovery (physical-to-physical) is slow, complex, and costly.

 Recovery of one VM by another eliminates OS and agent installation, reducing recovery time by about
U

60%.
VT

 VM encapsulation enables faster disaster recovery.

 Cloning VMs on remote servers offers an effective recovery solution.
 Only one clone VM is active; others stay suspended until needed.
 Cloud control centers activate the clone VM in case of failure.
 Snapshots allow live migration with minimal downtime.
 Only updated data and states are sent to keep clone VMs current.
 Recovery Point Objective (RPO) and Recovery Time Objective (RTO) depend on snapshot frequency.
 Security must be maintained during live VM migration.

4.3.4 Architectural Design Challenges

There are six open challenges in cloud architecture development.These challenges are identified by
Armbrust et al. as both obstacles and opportunities.

Dept. of CSE, SVIT Page 26

 4.3.4.1 Challenge 1—Service Availability and Data Lock-in Problem

 Relying on a single cloud provider creates a single point of failure.

 Using multiple cloud providers improves high availability (HA) and fault tolerance.
 DDoS attacks pose threats to cloud service availability.
 Some utility computing services help SaaS providers scale quickly to defend against DDoS.
 Interoperability among cloud platforms has improved, but APIs are still proprietary.
 Proprietary APIs cause data lock-in, making it hard to move data or applications.
 Standardizing APIs allows easier migration across cloud providers.

.IN
 Standard APIs support both public and private cloud usage (hybrid model).
 This enables "surge computing" — using public cloud resources when private cloud capacity is exceeded.
C
4.3.4.2 Challenge 2—Data Privacy and Security Concerns
N
 Public cloud networks are more exposed to attacks than private networks.
SY

 Existing technologies like encrypted storage, VLANs, and firewalls can help secure cloud systems.
 Encrypting data before storing it in the cloud adds protection.
 Some laws require SaaS providers to store data within national borders.
U

 Traditional network attacks include DoS, buffer overflows, spyware, and malware.
 Cloud-specific threats include hypervisor malware, guest VM hopping, and VM rootkits.
VT

 VM migrations are vulnerable to man-in-the-middle attacks.

 Passive attacks focus on stealing data or passwords.
 Active attacks can damage servers by altering kernel data structures.

4.3.4.3 Challenge 3—Unpredictable Performance and Bottlenecks

 Multiple VMs can share CPU and memory easily, but sharing I/O causes performance issues.
 Example: 75 EC2 instances show good memory bandwidth (1,355 MB/s) but poor disk write bandwidth (55
MB/s).
 I/O interference occurs when VMs compete for disk access.
 A solution is to improve I/O architecture and OS support for virtualizing interrupts and I/O channels.
 Internet applications are increasingly data-intensive and distributed across cloud boundaries.
 This complicates data placement and transfer, increasing cost and latency.

Dept. of CSE, SVIT Page 27

  Effective data traffic and placement strategies are essential to avoid bottlenecks.
 Amazon’s CloudFront is one example addressing these data transfer challenges.
 Bottleneck links should be widened and underperforming servers removed to improve overall performance.

4.3.4.4 Challenge 4—Distributed Storage and Widespread Software Bugs

 Cloud application databases are constantly growing and need scalable storage solutions.
 The goal is to design distributed SANs that scale up or down on demand.
 Data centers must support scalability, durability, and high availability (HA).

.IN
 Ensuring data consistency in SAN-connected cloud data centers is a significant challenge.
 Large-scale distributed bugs are hard to reproduce, requiring debugging in live production environments.
 Data centers typically do not allow in-production debugging.
C
 Virtual machines (VMs) can help capture valuable debugging information.
 Well-designed simulators offer another method for debugging distributed systems.
N
SY

4.3.4.5 Challenge 5—Cloud Scalability, Interoperability, and Standardization

 Storage and network bandwidth are billed based on the number of bytes used.
U

 Computation charges vary by virtualization level; GAE charges by CPU cycles and auto-scales with load,
while AWS charges hourly per VM instance, even if idle.
VT

 There is an opportunity to scale resources up and down rapidly based on load variation to save costs without
violating SLAs.
 Open Virtualization Format (OVF) provides a secure, portable, and platform-independent way to package
and distribute VMs.
 OVF supports packaging software to run on any virtualization platform, regardless of host or guest OS.
 It includes transport mechanisms for VM templates and supports multi-VM virtual appliances.
 Research is needed to create hypervisor-agnostic VMs and enable live migration between x86 Intel and
AMD systems.
 Further efforts are required to support legacy hardware and achieve effective load balancing across
heterogeneous platforms.

Dept. of CSE, SVIT Page 28

 4.3.4.6 Challenge 6—Software Licensing and Reputation Sharing

 Many cloud providers initially use open source software due to commercial software licensing not fitting
utility computing well.
 There is an opportunity for open source to stay popular or for commercial vendors to adapt licensing models
for cloud use.
 Combining pay-for-use and bulk-use licensing schemes can broaden business reach.
 One customer's bad behavior can damage the cloud's reputation, such as EC2 IP blacklisting by spam-
prevention services affecting VM installation.

.IN
 Creating reputation-guarding services, similar to trusted email services, could protect cloud providers and
users.
 Legal liability transfer between cloud providers and customers is a challenge that must be addressed in
C
SLAs.
N
4.4 PUBLIC CLOUD PLATFORMS: GAE, AWS, AND AZURE
SY

4.4.1 Public Clouds and Service Offerings

U

 Cloud services are requested by IT admins, software vendors, and end users.
 Five levels of cloud players exist, with individual and organizational users at the top level demanding
VT

different services.
 SaaS application providers mainly serve individual users.
 IaaS and PaaS providers primarily serve business organizations.
 IaaS offers compute, storage, and communication resources to both applications and organizations.
 PaaS providers define the cloud environment and support infrastructure services and organizational users.
 Cloud services depend on advances in virtualization, SOA, grid management, and power efficiency.
 Consumers buy cloud services as IaaS, PaaS, or SaaS.
 Many entrepreneurs offer value-added utility services to large user bases.
 The cloud industry grows as enterprises outsource computing and storage to professional providers.
 Provider service charges are usually much lower than the cost of frequent server replacements.
 Table 4.5 summarizes profiles of five major cloud providers as of 2010.

Dept. of CSE, SVIT Page 29

 .IN
C
4.4.2 Google App Engine (GAE)
N
 Google runs the world’s largest search engine infrastructure.
 They have expertise in processing massive amounts of data.
SY

 This led to innovations in data-center design and scalable programming models like MapReduce.
 Google operates hundreds of data centers worldwide with over 460,000 servers.
 Around 200 data centers run cloud applications simultaneously.
U

 Data stored includes text, images, and videos, all replicated for fault tolerance.
VT

 Google’s App Engine (GAE) is a PaaS platform for various cloud and web applications.

4.4.2.1 Google Cloud Infrastructure

 Google has led cloud development by using many data centers worldwide.

 Popular cloud services from Google include Gmail, Google Docs, and Google Earth, all supporting many users
with high availability.

 Google’s major technologies include Google File System (GFS), MapReduce, BigTable, and Chubby.

 In 2008, Google introduced Google App Engine (GAE), a platform for scalable web applications used by many
smaller cloud providers.

 GAE runs applications on Google’s extensive data center network linked to its search infrastructure.

 Key Google cloud components:

Dept. of CSE, SVIT Page 30

  GFS for storing large data sets,
 MapReduce for developing applications,
 Chubby for managing distributed locks,
 BigTable for structured and semi-structured data storage.

 Google’s data centers have thousands of servers organized into clusters running these services.

 GAE hosts third-party applications, so developers don’t manage servers directly.

 GAE’s frontend is an application framework like ASP or J2EE, supporting Python and Java.

4.4.2.2 GAE Architecture

.IN
 The Google cloud platform’s main components include Google File System (GFS) for large data storage,
MapReduce for application development, Chubby for distributed lock services, and BigTable for structured
C
data storage.
 These technologies are used together inside Google data centers, which contain thousands of servers
N
organized into clusters.
SY

 Users access Google applications through web interfaces; third-party developers can build cloud apps using
Google App Engine (GAE).
 Google’s core infrastructure is private and not open for external service building.
U

 GAE runs third-party applications on Google’s infrastructure, removing the need for developers to manage
VT

servers.
 GAE combines several software components, with a frontend framework similar to ASP, J2EE, or JSP.
 GAE supports Python and Java environments, functioning like web application containers and providing full
web technology support.

Dept. of CSE, SVIT Page 31

 .IN
C
4.4.2.3 Functional Modules of GAE
N
 GAE helps you build apps but doesn’t give you control of the actual servers.
SY

 It supports only Python, Ruby, and Java programming languages.

 You can use GAE for free with limits, and pay if you use more.
 Google takes care of all the hard server stuff for you.
U
VT

1. Datastore: Stores data safely using Google’s BigTable system.

2. App Runtime: Lets you build and run web apps using Python or Java.
3. SDK: A tool you use to create and test apps on your own computer before uploading them.
4. Admin Console: Helps you manage your apps easily.
5. Web Service Infrastructure: Manages storage and network smoothly for your apps.

4.4.2.4 GAE Applications

 Popular Google apps like Search, Docs, Earth, and Gmail run on GAE and support many users at once.
 Users access these apps through web browsers.
 Third-party developers can use GAE to build their own cloud apps.
 These apps run on thousands of servers inside Google’s data centers.
 GAE offers storage services for apps to save data securely and perform database-like operations (queries,
sorting, transactions).

Dept. of CSE, SVIT Page 32

  It also provides Google services like Gmail login, so developers don’t have to build their own user
management.
 Apps on GAE can easily authenticate users and send emails using Google accounts.

4.4.3 Amazon Web Services (AWS)

 Amazon AWS offers public cloud services mainly using the Infrastructure-as-a-Service (IaaS) model.
 EC2 provides virtual machines (VMs) where cloud apps run.
 S3 is Amazon’s object storage service for storing data like files.
 EBS offers block storage that works like a hard drive for traditional apps.

.IN
 SQS (Simple Queue Service) ensures reliable message delivery between processes, even if one is offline.
C
N
SY
U
VT

 Users access AWS services via standard protocols like SOAP through browsers or client programs.
 AWS offers many services across different application areas (12 tracks summarized in a table).
 AWS also provides SQS and SNS for messaging and notifications.
 ELB (Elastic Load Balancer) distributes incoming app traffic across EC2 instances to balance load and
avoid failing servers.

Dept. of CSE, SVIT Page 33

  Auto Scaling and ELB are managed by CloudWatch, which monitors resource usage and performance
(CPU, disk, network, etc.).
 AWS offers a Relational Database Service (RDS) similar to Azure’s.
 Elastic MapReduce is AWS’s big data service, similar to Hadoop, running on EC2.
 AWS Import/Export lets users physically ship disks to transfer huge data volumes quickly between distant
locations.
 Amazon CloudFront is a content delivery network to distribute content efficiently worldwide.
 AWS is flexible and popular with small and medium businesses to build and run cloud applications,
supporting large user bases and paid services.

4.4.4 Microsoft Windows Azure

.IN
C
 Microsoft launched Windows Azure in 2008 as a cloud platform built on Windows OS and Microsoft
virtualization.
N
 Azure runs applications on virtual machines (VMs) hosted in Microsoft data centers.
SY

 Azure manages all data center resources: servers, storage, and networks.
 The platform has three main components and provides various cloud-level services:
 Live Service: lets users access Microsoft Live apps and work on data across multiple machines.
U

 .NET Service: supports app development locally and execution on the cloud.
VT

 SQL Azure: cloud-based relational database service using Microsoft SQL Server.
 SharePoint Service: platform to build scalable business web applications.
 Dynamic CRM Service: platform to build and manage customer relationship management (CRM) apps for
finance, marketing, sales, etc.
 Azure services integrate well with other Microsoft apps like Windows Live, Office Live, Exchange Online,
SharePoint Online, and Dynamic CRM Online.
 It uses standard web protocols like SOAP and REST for communication.
 Users can integrate Azure cloud apps with other platforms and third-party clouds.
 The Azure SDK allows developers to build and test Azure apps locally on Windows before deploying to the
cloud.

Dept. of CSE, SVIT Page 34

 .IN
C
N
SY

4.5 INTER-CLOUD RESOURCE MANAGEMENT

4.5.1 Extended Cloud Computing Services

U

 Six layers of cloud services (from bottom to top):

VT

 Hardware as a Service (HaaS): Physical hardware resources like servers.

 Network as a Service (NaaS): Networking components connecting hardware (includes Virtual LANs).
 Location as a Service (LaaS): Data center space, power, and security for hardware; sometimes called
Security as a Service.
 Infrastructure as a Service (IaaS): Compute, storage, and communication resources. Can be divided
further into Data as a Service (DaaS) and Communication as a Service (CaaS).
 Platform as a Service (PaaS): Platforms for developing and running applications, built on IaaS.
 Software as a Service (SaaS): Software applications delivered over the cloud, built on PaaS.

 Layer dependency:

 SaaS depends on PaaS, which depends on IaaS, which depends on the lower physical layers. You cannot
run SaaS without having the underlying infrastructure in place.

Dept. of CSE, SVIT Page 35

  Cloud players divided into three groups:

 Cloud service providers and IT admins.

 Software developers or vendors.
 End users or business users.

 Different views on cloud models:

 Software vendors care most about application performance on the cloud platform.
 Providers focus on the cloud infrastructure itself.

.IN
C
N
SY
U
VT

4.5.1.1 Cloud Service Tasks and Trends

The five layers of cloud services and some examples:

1. Five layers of cloud services:

o SaaS (Software as a Service): Top layer with business apps like CRM (Customer Relationship
Management), collaboration tools, finance, and HR management. CRM was the first successful
SaaS.
o PaaS (Platform as a Service): Platforms to develop apps, offered by Google, Salesforce.com,
Facebook, etc.
o IaaS (Infrastructure as a Service): Cloud infrastructure like servers and storage, offered by
Amazon, Windows Azure, Rackspace, etc.
o Collocation services: Physical hosting and support across multiple cloud providers, often for
manufacturing supply chains.

Dept. of CSE, SVIT Page 36

 o Network cloud services: Communication services provided by companies like AT&T, Qwest,
AboveNet.
2. Vertical cloud services:
o These are connected, mutually supportive cloud services often combined together (called cloud
mashups) for specific industries or workflows.
3. Cloud service growth:
o Cloud services, especially SaaS for business applications, are rapidly growing.

4.5.1.2 Software Stack for Cloud Computing

 Cloud software is built in layers, each with a specific job.

 Lower layers like the OS are often customized for cloud needs.

.IN
 Cloud systems must be fast, always available, and handle failures well.
 Cloud platforms run on physical servers or virtual machines (VMs).
C
 VMs make cloud platforms flexible and not tied to specific hardware.
 The bottom layer stores huge amounts of data like a file system.
N
 Above storage are layers for databases, programming, and data queries.
SY

 Each layer talks to the one above to help build cloud applications.

4.5.1.3 Runtime Support Services

U

 Cloud clusters use cluster monitoring to check the status of all nodes.
VT

 A scheduler manages and assigns tasks to nodes based on availability.

 Cloud schedulers support special programming styles like MapReduce.
 Runtime support software helps keep the cloud system running smoothly and efficiently.
 In the SaaS model, users access software via browsers without buying or installing it.
 SaaS users don’t need to invest in servers or software licenses upfront.
 Cloud providers save costs compared to traditional hosting.
 User data is stored in cloud servers, which may be private or public clouds.

Dept. of CSE, SVIT Page 37

 .IN
C
N
SY
U
VT

Dept. of CSE, SVIT Page 38