CMPE 344 Computer Networks

Spring 2024

Foundations

Reading: Peterson and Davie, §1.1-1.5

Sources of slides:
Peterson and Davie, Computer Networks: A Systems Approach, 6th ed., Morgan Kaufmann, 2021.
Tanenbaum, Feamster, and Wetherall, Computer Networks, 6th ed., Pearson, 2021.
Kurose and Ross, Computer Networking, 8th ed., Pearson, 2021.

24/02/2024 1
 Goals of CMPE 344
• Emphasize on design choices made when
building computer networks
• Understand why networks are designed the way
they are
• Discuss “systems” approach to understand the
big picture
• Gain insight on architectural implications and
interaction of components rather than focusing
on rigidly defined layers

2
 Network Applications
• Most people are familiar with network through
applications:
– World Wide Web
– Email
– Online social network
– Streaming audio/video
– Videoconferencing (e.g. SkypeTM)
– File sharing
– Instant messaging
– … and many more
3
Network Applications

4
 Links and nodes
• A link is a physical medium that connects
computers or nodes.
– A node is a piece of hardware such as a PC,
a switch, a router, etc….
• Key challenge in network design: Scalability
– A system that is designed to support growth to
an arbitrarily large size is said to scale
– How should we connect all the nodes in the
world?

5
 Connectivity: Direct links

(a)

(b)

(a) Point-to-point
(b) Multiple access (Sharing a single physical link)
6
Connectivity: Switched network

• Connectivity between two nodes does not

necessarily imply a direct physical connection
between them
• Nodes can be indirectly connected: A switched
network contains forwarding nodes or switches
7
 Types of switched networks
• Circuit-switched networks
– Telephone networks
– A dedicated circuit across a sequence of links
is established. Source node sends a stream
of bits across this circuit to a destination node
• Packet-switched networks
– Computer networks
– Discrete blocks of data are sent in a (usually)
store-and-forward manner
8
 Packets
• Nodes in packet switched networks send
discrete blocks of data to each other
– These blocks correspond to some piece of
application data such as a file, a piece of
email or an image
– The blocks are called packets or messages
(more on this later)

9
 Store-and-forward
• In packet switched networks, each node
– first receives complete packet over some link
– stores the packet in its internal memory
– and then forwards the complete packet to the
next node
• Primary function of packet switches is to store
and forward packets

10
 Interconnection of networks

An internetwork or an internet

• A node that connects two or more networks is called a

router or gateway
• The Internet is an example of an internet 11
 Addressing and routing
• Each node in a network must have an identifier called its
address (a byte string)
• When a source node wants the network to deliver a
message to a certain destination, it must specify the
address of the destination node
• If the sending and receiving nodes are not directly
connected, switches and routers use the address to
decide how to forward the message toward the
destination
• The process of determining systematically how to
forward messages toward the destination node based on
its address is called routing

12
 Unicasting, broadcasting, and
multicasting
• Unicast: Source sends a message to a single destination
node
• Broadcast: Source sends a message to all the nodes on
the network
• Multicast: Source sends a message to some subset of
the other nodes (but not all of them)
• Thus, in addition to node-specific addresses, a network
may have to support broadcast and multicast addresses
as well
• Other “cast”s:
– Anycast: One-to-one-of-many (Applications?)
– Convergecast: Many-to-one (Applications?) 13
Cost-effective resource sharing
• How do hosts share the network and the same
link when they all want to use it at the same
time?
• Multiplexing: Sharing a system resource among
multiple users

L1 R1

L2 R2
Switch 1 Switch 2

L3 R3

Multiplexing/demultiplexing 14
 TDM, FDM, CDMA
• Synchronous time-division multiplexing (STDM) or just
time-division multiplexing (TDM)
– Time is divided into equal-sized quanta and each user
is given a chance to send data in a round-robin manner
• Frequency-division multiplexing (FDM)
– Each source transmits at a different frequency
• Code-division multiple access (CDMA)
– Each source has its own “code”
– Suitable for bursty data (we’ll define “bursty” later)
– In fact, it is a “multiple access” method (see more later)
Q: Find out what kind of multiplexing is used in cellular networks
and radio/TV broadcasting.
Q: What kind of multiplexing is used in fiber optic communications?
Hint: Optical carriers can be described by their wavelengths. 15
 TDM vs. FDM
Example:
FDM
4 users

frequency

time
TDM

frequency

time 16
 Limitations of TDM and FDM
• If one of the pairs does not have any data to
send, its share of physical link (time quantum or
frequency) remains idle even if one of the other
pairs has data to transmit
• In computer communications the idle time can
be large! (Q: Why?)
– We say that computer communications traffic is
bursty.
• Also, the maximum number of flows in these
schemes is fixed ahead of time
17
 Statistical Multiplexing
• Not all of the pairs are active at the same time!
• Like STDM: Link is shared over time
• Unlike STDM: Data is transmitted from each
user on demand rather than in a predetermined
slot
• Statistical multiplexing is efficient!

Q: How does one ensure that transmission decisions are made fairly?

18
 More on statistical multiplexing
• A packet switched network fragments a large
message into packets (that have a maximum
size) so that other flows can have a turn to
transmit their own
• Receiver will have to reassemble the packets
back into original message

Multiplexing packets
■■■
onto a shared link

19
 More on statistical multiplexing
• The switch may have to buffer packets in its
memory if it receives packets faster than the
shared link can handle
• If switch receives packets faster than it can send
them for an extended period of time, resource
will be congested
– Packets will be dropped
– Packets will be delayed excessively

20
Statistical multiplexing vs. circuit
switching
• Suppose users share a 1 Mbps link and each user
alternates between periods of activity (generates data at
constant rate 100 kbps) and periods of inactivity
• Further, each user is active 10% of time
• With circuit switching (TDM, FDM), 100 kbps must be
reserved for each user at all times
– E.g., TDM: 1-sec. frame is divided into 10 time slots of
100 ms each
– The link can support only 10 simultaneous users

21
 Statistical multiplexing vs. circuit
switching
• With statistical multiplexing, the probability that each user is active is
0.1 (10%)
• If there are 35 users, the probability that there are 11 or more
simultaneously active users is approx. 0.0004 (see next slide)
• When there are 10 or fewer simultaneously active users (which
happens with probability 0.9996), the aggregate arrival rate of data
is less than or equal to 1 Mbps (the output rate of the link)
• When there are more than 10 active users, queue will begin to grow
(until aggregate rate falls below 1 Mbps)
• The probability of having more than 10 simultaneously active users
is miniscule, the performance is the same as circuit switching, and 3
times the number of users are allowed

22
Statistical multiplexing vs. circuit
switching
• Let p=0.1
• Probability of having n simultaneously active
users out of 35 at any given time is

pn(1-p)35-n
• Probability that there are 11 or more users
transmitting simultaneously is
ଵ଴ ௡ ଷହି௡
௡ୀ଴

23
 Support for common services
• Goal: Hide the complexity of the network from
the application
• Logical channels: Application-to-application
communication path or a pipe

 Processes communicating
over an abstract channel
 Channel must meet the
functionality requirements
of applications

24
 Common communication
patterns
• After understanding the communication needs of a
representative collection of applications, extract their
common communication requirements, and finally
incorporate the functionality that meets these
requirements in the network
• Examples: Request/reply channels and message stream
channels
– Can you give example applications that make use of
these channels?

25
 Characterization of networks
according to their sizes
PAN Personal area network Around an
individual
SAN Storage area network In a room

LAN Local area network ~ 1 km

MAN Metropolitan area network ~ 10 km

WAN Wide area network Worldwide

26
 Three types of failures
• Bit errors
– 1 is turned into a 0 or vice versa
– Causes: Lightning, power surges, microwave ovens
– These are rare: 1 out of 106-107 on copper cable,
1 out of 1012-1014 on fiber
– Single bit errors and burst errors
• Packet losses
– Uncorrectable bit errors or packet drops at intermediate nodes
due to lack of buffer space
• Node- or link-level failures
– Crashed computers, broken links, misconfiguration
– Q: How do you distinguishing between a failed computer and
one that is merely slow? What about the difference between
between one that has been cut and one that is very flaky and
therefore introduces a high number of bit errors? (See more in
CMPE 449 Distributed Systems) 27
 Layering
• A fundamental design and implementation
concept
• When designing and analyzing complex
systems, we usually abstract away the details of
components and provide an interface for other
components of the system
• Services provided at higher layers are
implemented in terms of services provided by
lower layers

28
 Protocols
• Protocols
– are abstract objects that make up the layers of a
network system
– provide communication service that higher level
objects (e.g. application processes or higher level
protocols) use to exchange messages
– In fact, the term “protocol” is “overloaded”:
Specification of interface or module that implements it
• Two different interfaces are generally provided
– Service interface: To other objects on the same
computer
– Peer interface: To a protocol’s counterpart (peer) on
another machine 29
Service and peer interfaces of
protocols

30
 Encapsulation
• A header (and/or trailer) is attached to a
message body or payload
• Referred to as multiplexing and demultiplexing
up and down the protocol graph
• Headers attached to messages contain an
identifier that records the higher level object to
which the message belongs
• Trailers may contain information for error
detection/correction

31
 Encapsulation (cont.)

Messages
can be broken
down into smaller
messages

Encapsulation/decapsulation in an example 5-layer architecture 32

 Network architecture
• The set of rules governing the form and context
of a protocol graph is called a network
architecture
– The suite of protocols that make up a network
system is represented by a protocol graph
• International Standards Organization (ISO) and
Internet Engineering Task Force (IETF) are
examples of standardization bodies for network
architectures

33
Open Systems Interconnection
(OSI) reference model
End host End host

Application Application

Presentation Presentation

Session Session

Transport Transport

Network Network Network Network

Data link Data link Data link Data link

Physical Physical Physical Physical

One or more nodes

within the network 34
 Description of layers
• Physical Layer
– Handles the transmission of raw bits over a communication link
• Data Link Layer
– Collects a stream of bits into a larger aggregate called a frame
– Network adaptor along with device driver in OS implement the
protocol in this layer
• Network Layer
– Handles routing among nodes within a packet-switched network
– Unit of data exchanged between nodes in this layer is called a
packet

The lower three layers are implemented on all network nodes

35
 Description of layers
• Transport Layer
– Implements a process-to-process channel
– Unit of data exchanges in this layer is called a message
• Session Layer
– Provides a name space that is used to tie together the potentially
different transport streams that are part of a single application
– Example: Management of an audio stream and a video stream
that are being combined in a teleconferencing application
• Presentation Layer
– Concerned about the format of data exchanged between peers
– Is an integer is 16, 32, or 64 bits long? Is the most significant
byte transmitted first or last? How is a video stream formatted?
• Application Layer
– Standardize common type of exchanges
The transport layer and the higher layers typically run only on end-
hosts and not on the intermediate switches and routers 36
 Internet Architecture
FTP HTTP NV TFTP Application
TCP UDP
IP
TCP UDP Network

IP
Reference model:
No strict layering
NET 1 NET 2 NET n
■■■

Protocol graph: The hourglass shape

• Evolved out of ARPANET (DARPA/US DoD)

• Significance of the hourglass: Minimal and carefully chosen
set of global capabilities that allow many higher layer
applications and lower layer communication technologies to
coexist, share capabilities, and evolve rapidly 37
 IETF Internet architecture
• Main features
– Does not imply strict layering. The application is free to bypass
the defined transport layers and to directly use IP or other
underlying networks
– An hour-glass shape – wide at the top, narrow in the middle and
wide at the bottom. IP serves as the focal point for the
architecture (Thin waist: Minimalist approach): Common method
for exchanging packets among a wide collection of networks and
arbitrarily many transport protocols, each offering a different
channel abstraction to application programs
– In order for a new protocol to be officially included in the
architecture, there needs to be both a protocol specification and
at least one (and preferably two) representative implementations
of the specification
38
 The IP hourglass
Young Internet!

HTTP SMTP RTP …

QUIC DASH
Internet’s "thin waist”: many protocols
 one network layer TCP UDP
protocol: IP
in physical, link,
 must be IP transport, and
implemented by application
Ethernet PPP …
every (billions) of PDCP WiFi Bluetooth layers
Internet‐connected
devices copper radio fiber

Slide from Kurose and Ross, 8th ed.

The IP hourglass, at middle age
The Internet is about 40 years old!

HTTP SMTP RTP …

QUIC DASH

Internet’s middle age TCP UDP

“love handles”? caching
 middleboxes, IP
Firewalls
operating inside the
Ethernet PPP …
network PDCP WiFi Bluetooth

copper radio fiber

Slide from Kurose and Ross, 8th ed.

 Implementing network software
• Network protocols are implemented as part of the
operating system (OS)
• OS provides application programming interface (API) to
network services
• Socket interface
– Socket: Point where a local application process
attaches to the network
– Socket operations: Creating a socket, attaching the
socket to the network, sending/receiving messages,
closing the socket

41
 Network performance
• Network performance is measured in two fundamental
ways
– Bandwidth (or sometimes throughput): Number of bits
that can be transmitted over the network in a certain
period of time (e.g., 10 Mbps: i.e. it takes 0.1 µs to
transmit each bit)
• Notation: KB = 210 bytes; Mbps = 106 bits per second
– Latency (or delay): Time it takes a message to travel
from one end of a network to the other (e.g., 24 ms in
transcontinental networks)
• Round-trip time (RTT): Time it takes to send a message from
one end of a network to the other and back
42
 Bandwidth

(a)

(b)

(a) Bits transmitted at 1 Mbps

(b) Bits transmitted at 2 Mbps

We can talk about the bandwidth of a single physical link or

the bandwidth of a logical process-to-process channel
43
 Bandwidth vs. throughput
• In networking, it is customary to refer to bits-per-second
(bps) as the bandwidth
• Depending on the context (communication theory, signal
processing), bandwidth also refers to the range of
signals that can be accommodated
– e.g., Voice-grade telephone line, 300-3300 Hz, has a
bandwidth of 3000 Hz
– Relationship between bps and Hz will be established
in Chapter 2, Shannon’s Theorem
• Throughput is the bits-per-second (bytes-per-second,
packets-per-second, etc.) that we can actually transmit in
practice (refers to measured performance)

44
 Delay and throughput
• In general, latency has 3 main components
– Latency = Propagation + Transmit + Queue
– Propagation = Distance / SpeedOfLight
– Transmit = Size (bits) / Bandwidth (bps)
• Be careful since delay/latency/RTT are context
dependent
– We will make it explicit whenever necessary
• Effective end-to-end throughput
– Throughput = TransferSize (bits) / TransferTime (sec)

45
 Bandwidth vs. Latency
• Relative importance of bandwidth and latency
depends on application
– For large file transfer, bandwidth is critical
– For small messages (HTTP, NFS, etc.), latency is
critical
– Variance in latency (jitter) can also affect some
applications (e.g., audio/video conferencing) (see
more later)

46
Latency vs. bandwidth domination
10,000
5000

2000

1000

500

1-MB object, 1.5-Mbps link

200
1-MB object, 10-Mbps link
100 2-KB object, 1.5-Mbps link
2-KB object, 10-Mbps link
50
1-byte object, 1.5-Mbps link
1-byte object, 10-Mbps link
20

10

1
10 100
RTT (ms)
47
 Delay x Bandwidth product
• Volume of the pipe: The maximum number of
bits that could be in transit through the pipe at
any given interval
• A transcontinental channel with a one-way
latency of 50 ms and a bandwidth of 45 Mbps is
able to hold 50x10-3 sec x 45x106 bits/sec =
2.25x106 bits = 280 KB of data
Delay

Bandwidth

48
 Sample delay x bandwidth products
Link type Bandwidth Distance RTT Delay x BW

Dial-up 56 Kbps 10 km 87 µs 5 bits

Wireless 54 Mbps 50 m 0.33 µs 18 bits

LAN
Satellite 1 Gbps 35,000 km 230 ms 230 Mb

Cross- 10 Gbps 4,000 km 40 ms 400 Mb

country
fiber

Here, RTT = 2 * Propagation&processing delay 49

 Delay x bandwidth
• A network with a large delay-bandwidth product
is commonly known as a long fat network (LFN)
• Performance problems might arise when this
product is large
– “Large”: the product significantly exceeds 105
bits = 12,500 bytes
• Which links in the previous slide can be
considered as an LFN?

50
 Keeping the pipe full
• Delay x bandwidth is important to know when
constructing high-performance networks
because it corresponds to how many bits the
sender must transmit before the first bit arrives
at the receiver
• If the sender is expecting the receiver to
somehow signal that bits are starting to arrive
the sender can send up to 2 delay x bandwidth
(= RTT x bandwidth) worth of data before
hearing from the receiver
51
 Keeping the pipe full
• TCP will have performance problems when
there are paths with high bandwidth and long
round-trip delays (LFNs)
• The relevant parameter here is the product of
bandwidth (bits per second) and round-trip delay
(RTT in seconds)
– Number of bits it takes to "fill the pipe", i.e.,
the amount of unacknowledged data that TCP
must handle to keep the pipe full

52
 Time scales and performance
• Average vs. “instantaneous” bandwidth:
– Time interval over which average is computed is
important
– Suppose a video application needs 2 Mbps on
average:
• If it transmits 1 Mb in first second and 3 Mb in the following
second, over the 2-second interval average rate is 2 Mbps
• However, just knowing the average may not be enough.
What if the network can support no more than 2 Mb in any
one second
• Generally, one puts an upper bound on how large a “burst”
can be (Burst: Peak rate maintained for some period of time)

53
 Delay and jitter
• Delay requirement: As little delay as possible
• Jitter requirement: As little variation in delay as
possible
• Smoothing out jitter is important for better
performance in video applications
– Jitter can be smoothed out by delaying the
time at which playback starts
Interpacket gap

4 3 2 1 4 3 2 1
Packet Packet
source
Network sink

54
Network induces jitter!
 Cloudification or softwarization
of the network
• Network operators are trying to simultaneously accelerate the pace
of innovation (known as feature velocity) and yet continue to offer a
reliable service (preserve stability)
• Adopt the best practices of cloud providers:
– Take advantage of commodity hardware and move all
intelligence into software
– Adopt agile engineering processes that break down barriers
between development and operations
• Software Defined Networks (SDNs) is a game changer in terms of
how rapidly the network evolves to support new features (See more
in CMPE 448 Modern Networking Concepts)

55