COM 360

1
COM 360
2
Chapter 3

• Packet Switching

3
Limitations of Point-To-Point Networks

• Problem Not all Networks are Directly Connected
• Limit to number of hosts that can be attached
  (e.g. Ethernet 1024 maximum)
• Limit to distance a network can serve
• (e.g. Ethernet 2500 meters maximum)
• Next goal is to build networks that can be global
  in scale and to enable communication between
  hosts that are not directly connected.

4
Comparison to Phone System

• Phone is not connected to every person you might
  call- instead it is connected to an exchange with
  a switch.
• The switch creates the illusion that you are
  connected to the person at the other end of the
  call.
• Similarly computer systems use packet switches to
  enable packets to travel form one host to
  another, even when there is no direct connection
  between the hosts.

5
Packet Switches

• A Packet Switch is a device with several inputs
  and outputs leading to and from the hosts
  connected to the switch.
• Main role of the switch is to take packets that
  arrive on the input and forward (or switch) them
  to the right output for their destination.
• If packet arrival rates exceed the capacity of
  the output, the switch queues (or buffers) them.
  This is the problem of contention and the switch
  is said to be congested.

6
Switching and Forwarding

• A switch is a device that allows us to
  interconnect links to form larger networks.
• A switch is a multi-input, multi-output device,
  which transfers packets from an input to one or
  more outputs.
• A switch adds a star topology to the
  point-to-point link (Ethernet) and ring (802.5
  and FDDI)
• Topologies.

7
Topologies

• Networks can be classified by shape or topology.
  The most common are
• Bus
• Ring
• Star

8
Bus Topology

• Networks with a bus topology consist of a long
  cable to which the computers attach. Any computer
  connected to the cable can send a signal and all
  computers receive the message.

9
Ring Topology

• All computers are connected in a closed loop.
• Each computer is connected to exactly two others.
  
• The ring is a logical connection and the physical
  connections might look different and do not have
  to appear circular.

10
Star Topology

• All computers attach to a central point or hub

11
Star Topology in Practice

• Previous diagram is idealized usually,
  connecting cables run in parallel to computers

12
Reasons for Multiple Topologies

• Each topology has advantages and disadvantages
•  Star - protects the network from damage, since
  each cable connects only one computer.
•  Ring - makes connections and coordination
  easier, but if one cable is damaged or one
  computer crashes, the entire network is disabled.
• Bus - requires fewer wires than the star, but
  like a ring- if the main cable is damaged, the
  entire network is disabled.

13
A Switch Provides a Star Topology
14
Properties of a Star Topology

• Even though a switch has a limited number of
  inputs and outputs, large networks can be built
  by connecting switches.
• We can connect switches to each other and to
  hosts using point-to-point links allowing us to
  build large geographic networks.
• Adding additional hosts to a switch does not
  necessarily decrease the network performance.
• Every host on a switch has its own link to the
  switch and thus may transmit simultaneously.

15
Switching or Forwarding

• Switched networks are scalable or capable of
  growing to large numbers of nodes.
• Switching or forwarding is the main function of
  the network layer of the OSI Architecture.

16
Example Protocol Graph Running on a Switch
17
Example Switch for Previous Graph
With 3 input and 3 output ports
18
Switching

• How does a switch determine the port for a
  packet?
• It looks at the header for an identifier or
  address
• Datagram or connectionless approach
• Virtual circuit or connection-oriented approach
• Source routing less common
• Assumptions nodes must have globally unique
  identifiers and ports of each switch have
  identifiers (either numbers or names of
  connecting nodes)

19
Datagrams

• Idea behind a datagram is simple
• A datagram is the basic transmission unit in the
  Internet architecture.
• Datagram networks are connectionless.
• Every packet must contain a complete destination
  address
• Switch consults a forwarding or routing table
• Routing is a process by which nodes exchange
  information to build a routing table.

20
An Example Network
Datagram forwarding
21
Forwarding Table for Switch 2
22
Characteristics of Connectionless Datagram
Networks

• A host can send a packet anywhere, anytime, since
  switches immediately forward them.
• When a host sends a packet, it does not know if
  the network can deliver it.
• Each packet is forwarded independently of
  previous packets and possibly by different paths.
• A switch or link failure may not have serious
  effects since it may be possible to find an
  alternate route. This was a goal of the ARPANET.
• It was the ability to route around failures that
  led to a datagram based design ( especially
  important to the military).

23
Virtual Circuit Switching

• Connection oriented model uses a virtual circuit
  (VC)
• A widely used virtual circuit protocol is the
  Transmission Control Protocol (TCP).
• Switched virtual circuits (SVC) are generally set
  up on a demand basis and are disconnected when
  the call is terminated.
• A permanent virtual circuit (PVC) can be
  established as an option to provide a dedicated
  circuit link between two facilities.

24
Implementing A Virtual Circuit

• Two stage process
• Connection setup stage- establishes a connection
  between the source and destination hosts and
  creates a VC table in each switch.
• Data transfer stage- host puts an (VCI)
  identifier in the header and send it to the
  switch
• When the host no longer wants to send data it
  tears down the connection and the switch removes
  the relevant entries in the table.

25
Virtual Circuit Table

• An entry in a virtual circuit table for each
  switch contains
• A virtual circuit identifier (VCI)- that uniquely
  identifiers the connection
• An incoming interface
• An outgoing interface
• A potentially different VCI that will be used for
  outgoing packets.

26
Sending a Packet on a VC

• If a packet arrives on an incoming interface of
  the switch and that packet contains the
  designated VCI value in its header, then it is
  sent out on the outgoing interface with the
  specified VCI value now included in its header.
• Whenever a new connection is created, we need to
  assign a new VCI on each link and assure that the
  link is not currently in use by some existing
  connection.

27
Example of a Virtual Circuit
28
Virtual Circuit Tables
29
A Packet is Sent into a VC From A to B
A puts VCI value of 5 in header and sends to
switch 1. Switch 1 Uses the table and puts the
value 11in the header and sends it to Interface
3 on switch 2.
30
Packet Makes Its Way Though a VC
Switch 2 looks up the value in its VC table, puts
the value 7 in the header and send it out on
interface 2 to switch 3. This continues until the
packet arrives at B.
31
Virtual Circuits

• Note that the combination of the VCI of packets
  as they are received at the switch and the
  interface on which they are received uniquely
  identifies the virtual connection.
• The VCI has link local scope and is only
  significant for that connection
• When a new connection is created, a new VCI has
  to be assigned to the connection.
• Need to insure that the chosen VCI on a link is
  not currently in use by some existing connection.

32
Virtual Circuits

• To establish a connection
• The network administrator can configure the state
  in which case it is called permanent(PVC).
• A host can send messages into the network to
  cause the state to be established. This is
  referred to as signalling and the resulting
  circuits are said to be switched (SVC).

33
Virtual Circuit Switching

• By the time the host gets the go-ahead to send
  the data, it knows a lot about the network.
• The connection-oriented model does the following
• Allocates buffers to each virtual circuit
• Runs the sliding window protocol between each
  pair of nodes
• The circuit is rejected by the host if there are
  not enough buffers available
• Thus each node is ensured of having the buffers
  it needs. This is called hop-by-hop control.

34
Datagram Network

• There is no connection phase and each switch
  processes each packet individually.
• Each packet competes with others for buffer
  space. If there are no buffers, the packet is
  discarded.
• It is possible to distinguish among the packets
  to try to ensure that they receive a fair share
  of the buffers

35
Quality of Service

• Quality of Service (QoS) means that the network
  gives the user some kind of performance related
  guarantee, which implies that the switches set
  aside the resources to meet this guarantee.

36
Examples of Virtual Circuits

• Frame Relay- simple implementation -provides some
  basic quality of service and congestion avoidance
  features- used in the construction of virtual
  private networks (VPN).
• Asynchronous Transfer Mode (ATM)

37
Frame Relay Packet Format
Frame Relay packet format provides a good example
of a packet used for virtual circuit switching.
38
Contention and Congestion

• Contention occurs when multiple packets are
  queued at a switch because the are competing for
  the same output link.
• Congestion means that the switch has so many
  packets queued that it runs out of buffer space
  and has to start dropping packets.
• The Datagram model experiences congestion.

39
Source Routing

• Source Routing is a third approach to switching.
  All the information that is needed to switch a
  packet across the network is provided by the
  source host.
• One way to do this is to put an ordered list of
  switch ports in the header and to rotate the list
  so that the next port is always at the front of
  the list.
• (Not commonly used today.)

40
Source Routing In A Switched Network
Switch reads the rightmost-number
41
Source Routing

• Assumes that the host knows enough about the
  topology to form a header that has all the
  switches.
• We cannot predict how big the header may be,
  since it must be able to hold one word of
  information for every switch on the path.
• There are some variation of this approach
• Instead of rotating the header, the switch can
  strip off the element just used
• The header can carry a pointer to the next port
• Can be used in both datagram networks and virtual
  circuits, but it does not scale well.

42
Handling Headers For Source Routing
a) rotation
b) stripping
c) pointer
These labels are read from right to left.
43
Bridges and LAN Switches

• Originally repeaters were used to connect a pair
  of Ethernet segments.
• An alternative was to put a node between the two
  Ethernets and have the node forward frames. This
  node is called a bridge.
• Bridges just accept LAN frames and forward them
  on the outputs.
• This provides a way to increase the total
  bandwidth of the network. If a segment can carry
  10Mbps, a bridge can carry as much as 10n Mbps,
  for n ports on the bridge.

44
Learning Bridges

• Bridges do not have to forward all the frames it
  receives.
• It can learn on which port each host resides.
• A table can be downloaded and referred to when
  each new frame arrives.
• A bridge can learn this information by inspecting
  the frames it receives and updating its table.
  When the bridge boots the table is empty and
  entries are added over time. If a frame is not on
  the table it is forwarded to all.

45
A Learning Bridge
46
Forwarding Table Maintained By a Bridge
47
Spanning Tree Algorithm

• Extending LANS with Bridges works well until it
  has a loop in it, which can allow a frame to
  circulate forever.
• A loop can be introduced by an administrator,
  when a network spans multiple departments.
• Bridges run a distributed spanning tree
  algorithm, which is a sub-graph which keeps all
  of the original vertices and eliminates some of
  the edges.

48
Extended LAN With Loops
49
Spanning Tree With Some Ports Not Selected
B1 is the root node, B3 and B5 are connected to
LAN A and will use B5 since it is closer. B5 and
B7 are connected to LAN B. B5 is the designated
bridge because it has smaller ID and both
are equidistant.
50
Graphs and Spanning Trees
a) Cyclic graph
b) Corresponding spanning tree
51
Broadcast and Multicast

• Bridges forward unicast frames from one LAN to
  another.
• Goal of a bridge is to extend a LAN across
  multiple networks, and most LANS support both
  broadcast and multicast, bridges also need to
  support these.
• Broadcast- each bridge forwards a frame out on
  each active port
• Multicast- similar to broadcast and allows each
  host to decide whether or not to accept the frame.

52
Limitations of Bridges

• The bridge-based solution is used only to connect
  a limited number of similar LANS.
• Scale- cannot connect more than a few LANs with
  bridges.
• Homogeneity- bridges are fairly limited in the
  kinds of LANs they can connect. They make use of
  the network frame headers, which must have same
  format. Can connect Ethernet to Ethernet, 802.5
  to 802.5, etc.

53
Virtual LANs

• For scalability, LANs may be extended by using
  Virtual LANs (VLAN).
• VLANs allow a single extended LAN to be
  partitioned into several separate LANs, each with
  an identifier (called a color), and packets can
  only travel from one segment to another if both
  have the same identifier.
• This limits the number of segments that will
  receive a broadcast packet.

54
Virtual LANs With Common Backbone
Any broadcast packet will be received by any
host. Suppose W,X and Y,Z are two VLANs and B1,
B2 belong to both. A broadcast packet from X
will be sent only to W
55
Cell Switching (ATM)

• Asynchronous Transfer Mode became an important
  technology in the late 1980s and early 1990s.
• It was used by the telephone industry
• IT became available for use as a high speed
  switching technology, just when other shared
  media like Ethernet and X.25 were becoming too
  slow.

56
ATM

• A network technology based on transferring data
  in cells or packets of a fixed size.
• The cell used with ATM is relatively small
  compared to units used with older technologies.
• The small, constant cell size allows ATM
  equipment to transmit vide, audio, and computer
  data over the same network, and assure that no
  single type of data monopolizes the line.
• This differs from other technologies based on
  packet-switched networks (such as the Internet ot
  the Ethernet), in which variable sized packets
  (called frames) are used.

57
ATM

• ATM is a connection-oriented, packet switching
  technology, which uses virtual circuits.
• The connection set-up phase is called signaling.
• When a virtual connection is set up, the address
  of the destination is put in the signaling
  message, using one of several formats.
• ATM packets are of fixed length- 53 bytes (header
  and a 48 byte payload) and are called cells.

58
Cells

• Cells, in contrast to other packets, are fixed in
  length and small in size.
• Another property of cells relates to the behavior
  of queues. Fixed-length cells means that a queue
  output is not tied up for more time that it takes
  to transmit one cell.
• Queues of cells also tend to be shorter that
  queues of packets.

59
Cell Size

• Why use fixed size cells?
• Fixed packets made building fast, scalable
  switches easier because
• It is easier to build hardware to do simple jobs
  and processing packets is simpler when you know
  the length of each packet.
• If all packets are the same length, you can have
  many switching elements doing the same thing in
  parallel, and each taking the same time to do its
  job.
• Enabling parallelism, improves the scalability of
  switch design.

60
Cell size

• The motivation for the use of small data cells
  was the reduction of jitter(delay variance) in
  the multiplexing of data streams
• Reducing jitter and also end-to-end round-trip
  delays is particularly important when carrying
  voice traffic.

61
Cell Size

• Having decided to use small fixed-length packets,
  the next question is What is the right length
  for these cells?
• The resulting size was a compromise between the
  US proposed 64 bye size and the European 32 byte
  size. Unfortunately, it is not a power of 2.

62
Cell Format

• The ATM cell comes in two different formats
• UNI (User-network interface)- used between a
  telephone company and it customer.
• NNI ( Network-to-Network Interface) is used
  between pairs of phone companies.
• The primary difference is that the NNI format
  replaces the GFC(generic flow control) field with
  4 extra bits of VPI ( virtual path identifier).

63
ATM Cell Format
This is an example of the UNI cell format.
64
Segmentation and Reassembly

• Up until now we have assumed that a low-level
  protocol could accept the packets handed down
  form a higher level protocol, attach its header
  and pass the packet down.
• This is not possible with ATM, since the packets
  are often larger than 48 bytes and will not fit
  in the ATM payload.
• Solution Fragment the packet at the source.

65
Segmentation and Reassembly

• Up until now we have assumed that a low-level
  protocol could accept the packets handed down
  form a higher level protocol, attach its header
  and pass the packet down.
• This is not possible with ATM, since the packets
  are often larger than 48 bytes and will not fit
  in the ATM payload.
• Solution Fragment the packet at the source.

66
Segmentation and Reassembly

• The technique called fragmentation and reassembly
  is often called segmentation and reassembly (SAR)
  in the case of ATM.
• It involves fragmenting the packet, transmitting
  the individual packets and then reassembling the
  fragments back together at the destination.
• It is more of a problem with ATM, than with a
  network with a maximum packet size of 1500 bytes.
• An additional layer, the ATM Adaptation Layer
  (AAL) was added to handle this.

67
Segmentation and Reassembly in ATM
ATM supports many services, including voice,
video, and data, and its service have different
AAL needs. AAL3 is used by connection oriented
services (like X.25) and AAL4 is used by
connectionless services ( such as IP). The
relation between AAL and ATM is shown here
68
ATM Adaptation Layer 3/4

• Main function of AAL3/4 is to provide variable
  length packets to be transported across ATM
  networks.
• Supports segmentation and reassembly

69
ATM Adaptation Layer 3/4 Packet Format
70
ATM Cell format for AAL 3/4
71
Encapsulation and Segmentation for AAL3/4
72
ATM Adaptation Layer 5

• Replaces the type field of AAL3/4 with 1 framing
  bit in the ATM header, simplifying it.
• AAL5 is now preferred for transmitting IP
  datagrams over ATM
• Uses bandwidth more efficiently and it has a much
  simpler design than AAL3/4.

73
ATM Adaptation Layer 5 Packet Format
74
Encapsulation and Segmentation for AAL5
75
Virtual Paths

• ATM uses a 24 bit identifier for virtual
  circuits, and it is split into two parts
• 8 bit virtual path identifier (VPI) and
• 16 bit virtual circuit identifier (VCI)
• This creates two levels of connections
• Virtual path acts like a fat pipe that contains a
  bundle of virtual circuits

76
Example of a Virtual Path
77
Advantage of Virtual Path

• Although there may be thousands of virtual
  connections across the public network, the
  switches in the public network behave as though
  there is only one connection.
• This requires much less less connection-state
  information in the switches.

78
Physical Layers For ATM

• ATM usually runs on top of a SONET physical
  layer, although it can run over several different
  physical media.
• Main issue is how to find the boundaries of the
  ATM cells the framing problem.
• One of the SONET frame bytes can point into the
  SONET payload to the start of the ATM.

79
ATM in the LAN

• ATM is a switched technology, whereas Ethernet
  and 802.5 were shared-media technologies.
• ATM was designed to work with speeds gt 155Mbps,
  compared to the original Ethernet
• (10 Mbps) and token rings (4 or 6 Mbps)
• ATM switches have a performance advantage over
  shared media networks.
• ATM does not have distance limitations

80
ATM Used as a LAN Backbone
ATM became popular for the high-performance
backbone of larger LANs. Hosts were connected to
Ethernet switches, which were connected to ATMs.
81
Gigabit Ethernet

• Gigabit Ethernet links use the original framing,
  but are usually point-to-point fiber links and
  can run over longer distances (up to several
  kilometers).
• Same approach can scale up to 10 Gbps links.

82
Problems with ATM in a LAN

• ATM behaves differently than a shared media LAN,
  which supports broadcast and multicast.
• ATM can be made to behave like a LAN, called LAN
  emulation. This involves adding new protocols and
  addressing.
• LAN Emulation (LANE) adds functionality, through
  the addition of a number of servers.
• Devices that connect to an ATM network- hosts,
  bridges, routers are referred to as LAN emulation
  clients. (LEC)

83
LAN Emulation

• Servers that are required to build an emulated
  LAN are
• LAN emulation configuration server (LECS)
• LAN emulation server (LES)
• Broadcast and unknown server (BUS)
• These can be physically located in one or more
  devices
• The LECS and LES perform configuration functions
  and the BUS makes data transfer resemble that of
  a shared media LAN.

84
Protocol Layers in LAN Emulation
85
Servers and Clients in an Emulated LAN
86
Implementation and Performance

• There is a simple way to build a switch
• Buy a general purpose workstation and equip it
  with a number of NIC cards.
• Such a device with suitable software can receive
  packets on its interfaces and perform switching
  functions, and send packets out oon its
  interfaces.
• Very similar to low-end routers.

87
Workstation Used as a Packet Switch
Shows path of a packet which uses DMA. Problem
all packets pass through I/O buss twice.
Throughput it either ½ main memory bandwidth or ½
the I/O bus bandwidth, whichever is less.
88
Problem With Using Workstation As A Switch

• Main problem performance is limited by fact that
  all packets must pass through single I/O bus
  twice.
• Cost of processing a packet (parsing the header,
  deciding on which output link to transmit it)
  will dominate. ( Bad for short packets.

89
Example

• Suppose a workstation can perform processing to
  switch 500,000 packets per second (pps).
• If a packet is short (64 bytes) then
• Throughput pps x (bits per packet)
• 500 x 103 x 64 x 8
• 256 x 106
• which is a throughput of 256 Mbps much
  below todays expected throughput.

90
Defining Throughput

• If a switch has n inputs that each support a link
  speed of si the best possible throughput is the
  sum of all the si
• This is not achievable in practice, since a
  switch can only handle traffic arriving at full
  link speed on all inputs if it is evenly
  distributed.
• For Ethernet switches, the size of the packets
  also affects the switch performance.
• Throughput of the switch is a function of the
  traffic.

91
Traffic Modeling

• A traffic model approximates the behavior of real
  data traffic.
• It attempts to answer several important
  questions
• When do packets arrive?
• What outputs are they destined for?
• How big are they?

92
Ports

• The input or output on which packets are received
  or sent.
• A switch consists of input ports, output ports
  and a fabric.
• There is usually at least one control process
  that communicates with the ports either directly
  or via the switch fabric.
• The ports communicate with the outside world.
• Ports contain fiber optic receivers and lasers,
  buffers to hold packets and other circuitry.
• The fabric delivers the packet to the right
  output port.

93
Ports

• The ports maintain a list of virtual circuit
  identifiers that are in use.
• The input port receives a steady stream of
  packets and has about 200 nanoseconds to process
  a packet. (See p. 214)
• Ports also buffer packets on both input and
  output ports.
• Buffering in the fabric is called internal
  buffering.

94
A 4 x 4 Switch
An input buffer is often implemented as a FIFO.
Only one packet at a time can be forwarded to a
specific output Port, the rest remain in the
buffers and can prevent packets further back from
going to their Ports
95
Illustration of Head-Line Blocking
Notice that a packet destined for Port 1 is
blocked by a packet contending for Port 2. This
can limit throughput to 59 of its maximum and
should be avoided.
96
Fabrics

• A fabric is the part of a switch that actually
  does the routing. It should move packets with
  minimal delay.
• Shared bus- found in workstations used as a
  switch
• Shared memory- packets are read into memory by
  input port and read by output port
• Crossbar- a matrix of pathways configured to
  connect any input and output ports
• Self routing- rely on information in header to
  direct packets most scalable

97
A 4 x 4 Crossbar Switch
Main problem require each output port to accept
packets from all input ports at once.
98
Self- Routing

• Self-routing fabrics rely on information in the
  packet header to direct each packet to its
  correct output.
• A self-routing header is appended to the packet
  by the input port and is removed before the
  packet leaves the switch.

99
Self-Routing Header

• Packet arrives at input Port
• Input Port attaches header
• Self-Routing header is removed

100
Routing Packets Through a Banyon Network
Banyon Network is an Arrangement of
2x2 Switching elements that routes packets to
the correct Output without collisions if the
packets are presented In ascending order. It
uses the perfect shuffle wiring pattern
101
Summary

• Large scalable networks are built using switches.
• An important application of switching is the
  interconnection of shared-media LANs.
• Virtual Circuit switching is used in Frame Relay
  and ATMs.
• ATMs use cells, or short fixed length packets.
• Switches forward packets at a very high rate.

102
Open Issue Future of ATM

• Success of Ethernet switches has made ATMs less
  popular. Gigabit Ethernet and 10-Gigabit Ethernet
  have provided high speed connections to servers.
• Another factor limiting the use of ATMs is the
  Internet, with a service that delivers IP
  packets.
• ATMs are still used in Virtual Circuits but are
  being challenged by newer technologies.

103
Figure 3.33 for Ex. 1 and 2
104
Figure 3.34- Network for Ex. 3
Give datagram forwarding table for each node,
using lowest cost path
105
Figure 3.35- for Ex. 4
106
Figure 3.36Virtual Circuit Switches for Ex. 5
107
Figure 3.37 for Ex 13, 14
108
Figure 3.38 for Ex. 15, 16
109
Figure 3.39 for Ex. 17
110
Figure 3.40 for Ex. 18
111
Figure 3.41 for Ex 19, 20