The Network Layer

1
Lecture 5

• The Network Layer

2
Introduction

• The Network Layer is concerned with getting
  packets from the source to the destination
• It is the lowest layer that deals with end-to-end
  transmission
• It must know the topology of the communication
  subnet (the set of all routers)
• choose appropriate paths through the subnet
• When the source and the destination are in
  different networks, the network layer has to deal
  with the differences and solve the problems
  resulting from them

3
Network Layer Design Issues

• Services Provided to the Transport Layer
• The network layer provides services to the
  transport layer at the network layer/transport
  layer interface
• This is the interface between the carrier and the
  customer - the boundary of the subnet
• The carrier often has control of the protocols
  and interfaces up to and including the network
  layer
• The carrier deliver packets from the customers

• Two schools of thought
• Connectionless service (The Internet)
• the internet community and equipment vendors
• control, complexity, and intelligence in the
  hosts and the transport layer
• each packet with full destination address and
  independently routed
• usually argue for a speedy delivery
• argue that its cheaper and safer to put
  complexity in hosts
• Connection-oriented service (ATM)
• the carrier (service) providers
• control, complexity, and intelligence in the
  subnet and the network layer
• usually argue for an accurate delivery
• (TCP/)IP over ATM

4

• Internal Organization of the Network Layer
• virtual circuit (as contrast to fixed or
  permanent circuit)
• each packet traveling along a virtual circuit
  must contain the VC id
• VC entries are in the table of each router
• VC can be simplex or duplex (call setup can be
  from both directions)
• if two VCs (actually identical paths) share the
  same id then there can be a confusion
• datagrams
• each datagram must have the full destination
  address
• this address can be quite long
• Comparison of Virtual Circuit and Datagram
  Subnets
• datagrams
• if packets are short then the full destination
  address in datagrams is a big overhead
• where to route the datagram can be complicated
• congestion avoidance is more difficult
• virtual circuit
• the table space in routers is an overhead
• call setup time overhead especially credit card
  validation
• once setup routing is easy, just follow the vc

5
Routing Algorithms

• Session routing routing decisions are made only
  when a new vc is being setup and thus a route
  remains for an entire user session
• Nonadaptive algorithm routing decisions are
  made in advance, off-line, and not based on the
  current traffic and topology also called static
  routing
• Adaptive algorithms changes the routing
  decisions to reflect changes in the topology and
  traffic
• The Optimality Principle
• if the node J is on the optimal path from router
  I to router K, then the optimal path from J to K
  also falls along the same route
• sink tree
• no circuit since it is a tree
• the distance matrix is the number of hops

6

• Shortest Path Routing
• shortest in terms of
• the number or hops or
• the distance or
• queueing and transmission delay (in this case
  shortest means thefastest)
• Dijkstras algorithm
• from A to D
• label the nodes that are adjacent to A with the
  distance B(2,A) B(distance 2, from A)
• label the nodes that are not adjacent to A with
  infinity
• choose the one with the smallest label B in this
  case
• move to B and examine all adjacent nodes
• the nodes C(9,B) and E(4,B) are relabeled since
  9,4 lt infinity
• E(4,B) is chosen
• G(5,E) and F(6,E) are relabeled
• G(5,E) is chosen
• H(9,G) is relabled
• among C,F,H, F(6,E) is chosen
• H(8,F) is relabeled
• Floyds shortest path algorothm

7

• Flooding
• every incoming packet is sent out on every
  outgoing line except the one it arrived on
• creates vast numbers of duplicate packets
• can have a hop counter in the header of each
  packet
• set to a specific max number
• gets decremented at each hop
• with packet ids a node can keep the packet list
  that have been visited
• remove if visiting the second time
• selective flooding (heuristics)
• if flooding is from NYC to LA then at Chicago it
  is not sent to Boston direction lines
• very wasteful but robust
• military application
• distributed database to update all DBs in the
  network
• useful as a benchmarking tool for other routing
  algorithms
• flooding always chooses the shortest path (at
  least one of it will find it)

8

• Flow-Based Routing (static)
• given the following information routing can be
  precalculated
• the subnet topology
• the traffic matrix
• the line capacity
• a routing algorithm
• arrival for AB line is 14 packets/sec
• AB line capacity is 20kbps or 25 packets/sec
  (20,000 / 800 bit packet length)
• the packet service time is 1/800 1.25 msec
  (20,000 x 0.00125 25 packets/sec)
• AB transmission time is 1/(25-14) 0.091 sec
• the weight is the fraction of the total traffic
  using the line
• (integer) linear programming is often used to
  calculate the rightmost table

9

• Distance Vector Routing (dynamic)
• used in original ARPANET, in the Internet as RIP
  (Routing Information Protocol), DECnet, Novells
  IPX AppleTalk and CISCO routers use improved
  version of this
• J sends test packets to A, I, H, and K and
  measures the delay as 8, 10, 12, and 6
• A, I, H, K send their routing tables to J
• J recalculates each entry of its routing table
  with received/measured information
• the cheapest was from J to G is via H
• via H is 18 12 6
• via I is 41 10 31
• via K is 37 6 31
• The Count-Infinity Problem (b)
• (a) good news travels fast
• A coming up from down state
• (b) bad news travels slow
• A just went down
• B knows something went wrong with A
• but C says it can get there in 2 hops
• use the info B updates its table
• C relying on Bs table updates its table
• it propagates to D and E and B

10

• Distance Vector Routing (dynamic) - continues
• The Split Horizon Hack
• The Count to Infinity problem comes from the fact
  that B relies on Cs information to A where, in
  fact, Cs information was from B
• The solution is for C to tell B that to A is
  infinity Im relying on your input as far as A
  is concerned.
• D tells E that it can reach A in 3 hops but tells
  C that it is infinity, Im relying on your input
  to A.
• works except the case on the right
• A and B are the same distance away from D
• D goes down C reports to A and B
• A thinks it can still get to D via B by 3 steps
• B thinks it can still get to D via A by 3 steps
• A,B will update it to 4 steps
• it will be the same as the count to infinity
  problem

11

• Link State Routing
• after 1979, the Distance Vector Routing has been
  replaced by the Link State Routing for ARPANET
• DVR didnt take the bandwidth into account
• all lines were 56K but later to 230 kbps and
  1.544 Mbps
• DVR took long to converge with the split horizon
  hack problem
• Learning about the Neighbors
• upon booting each router learns from the
  neighbors as to who they are via HELLO packet
• Measuring Line Cost
• by sending ECHO packet the delay to its neighbor
  is measured
• Building Link State Packets
• either periodically or after changes with the
  neighbors (up or down) LSPs are built
• see figure (a) with the delay on each link (b)
  LSPackets A has connections to B and E
• flooding is used to send these packets
• Sequence number is the id of the packet
• Age is the number of seconds to live
• Distributing the Link State Packets
• different versions of routing tables can coexist
  in the network due to different arrival time
• seq and age are used
• the packet buffer indicates actions to take

12

• Computing the New Routes
• once a full set of link state packets are
  received each router can construct the entire
  subnet graph
• usually Djikstras shortest path algorithm is run
  for each pair of nodes
• Hierarchical Routing
• for a big network, the routers are divided into
  regions
• decrease in the table entries
• can increase the path length
• best route for 1A to 5C is via region 2 (5 hops)
  but
• all traffic to to region 5 is via region 3 (6
  hops)

13

• Routing for Mobile Hosts
• foreign agents (where the user is registering to
  use a service) and home agents (where the user is
  registered)
• deregistration is usually done by hanging up by
  user
• scenario of sending a packet to a mobile user
  registered with a foreign agent
• sender in Seattle sends a packet to the
  recipient's home agent (Seattle to NYC)
• the home agent looks up the temporary location
  and the foreign agents address the user is in
  LA now
• the home agent encapsulate the original packets
  data into another packet and sends the new packet
  to the foreign agent (Tunneling) (NYC to LA)
• the home agent sends the foreign agents info to
  the sender in Seattle
• the foreign agent takes out the original data and
  sends it to the mobile user
• a tunnel is established between the sender in
  Seattle and the user in LA for further
  communication

14

• Broadcast Routing
• send a distinct packet to each destination
• wasteful bandwidth
• need to know all the addresses
• flooding
• too many packets, too many bandwidth
• multidestination routing
• destination includes all the nodes to be
  delivered to
• routing node creates multiple packets with only
  those relevant addresses (like partition)
• after some hops the destination field will have
  only one address
• use sink tree (spanning tree)
• the information is available with link state
  routing but
• not available with distance vector routing
• reverse path forwarding (heuristic for spanning
  tree)
• each node pair has a preferred line for
  communication
• if a broadcast packet arrives it establishes a
  link of a tree
• a sink tree can be built
• reasonably efficient and easy to implement
• no need to keep the spanning tree information

15

• Multicast Routing
• the network, when huge, can be partitioned (not
  necessarily disjoint) into groups so that the
  communication can be directed to a certain group
  (multicast).
• A multiple smaller spanning trees can be built on
  a network.
• Storing many pruned spanning trees for different
  groups can be a problem. - core base tree
• a single spanning tree per group (instead of m
  possible spanning trees) is stored
• the root (the core) is usually in the middle of
  the group
• a multicast message is sent to the core
• may not be optimal for all sources but
• reduces the overhead of keeping m spanning trees

16
Congestion Control Algorithms

• Causes for congestion
• a sudden increase of input stream
• few slow processors in the network
• Congestion tends to get worse
• Congestion control is a global issue to the
  subnet involving all hosts and routers
• Flow control relates to point-to-point traffic
  between the sender and a given receiver
• Even if there is no congestion problem flow
  control may still be needed
• to give other hosts a chance to send

17

• General Principles of Congestion Control
• Open Loop solution
• to solve the problem by good design
• no midcourse correction
• act at source design
• act at destination design
• Closed Loop solution
• feedback loop like Neural Net
• explicit feedback
• implicit feedback
• Increase capacity
• certain services can temporarily request more
  bandwidth
• use backup bandwidth for emergency
• Decrease Traffic
• deny service to users
• degrading service to some or all users
• What if you cant increase capacity and cant
  decrease traffic?
• Service Outage.
• Stock price dives

18

• Congestion Prevention Policies
• the issues in the Transport layer are similar to
  those in the Data Link layer but
• the Transport layer handles the traffic across
  the whole network whereas
• the Data link layer handles between two routers
  over a wire

19

• Traffic Shaping
• widely used in ATM networks
• the main cause of congestion is that traffic is
  often burst
• Traffic Shaping is to force the hosts to transmit
  packets at a more predictable rate
• Carriers and customers agree in advance on the
  traffic pattern (shape)
• this agreement may not be so important for file
  transfers but very important for real time
  traffics like audio and video
• monitoring a traffic is called traffic policing
• monitoring traffic is easier on VC than datagrams
• The Leaky Bucket Algorithm
• leaky bucket discharges water at a constant rate
• any amount more than the bucket can take will
  overflow
• consider the leaky bucket as a finite internal
  queue
• once queue is full any additional
  traffic/arrivals will get discarded
• a single server queueing system with constant
  service time
• for a fixed packet size like ATM cells, a packet
  can be sent out at a constant rate
• for a variable length packets, a fixed number of
  byes can be sent out
• The Token Bucket Algorithm
• tokens are added to the bucket at every delta T
  time
• the same number of packets as the number of
  tokens can be sent out (or bytes)

20

• Flow Specification
• Traffic shaping or congestion control in general
  requires the agreement among the sender,
  receiver, and the subnet
• Such an agreement is called a flow specification
• Max packet size
• Token bucket rate (bytes/sec) rate of arrival
  into the bucket
• Token bucket size (bytes) the capacity of the
  bucket
• Maximum transmission rate (bytes/sec) shortest
  interval in which the token bucket could be
  emptied
• Loss sensitivity (bytes) / Loss Interval (micro
  sec) max acceptable loss rate
• Burst loss sensitivity (packets) how many
  consecutive lost packets can be tolerated
• Minimum delay noticed (micro sec) how long a
  packet can be delayed w/o the application
  noticing
• Maximum delay variation (micro sec) some
  applications are more sensitive to jitter (the
  amount of variation in the end-to-end packet
  transit time) than the delay synchronization
  issue
• Quality of guarantee states Are above
  specification just goals/wish list or absolute
  requirement? Some applications do not know what
  they want or what they mean. These
  specifications are negotiated between the Subnet
  and the users

21

• Congestion Control in Virtual Circuit Subnets
• What to do when congestion occurs?
• admission control (open loop at source) dont
  let any new request into the network
• detour the congested area see figure
• reservation after negotiation subnet set aside
  enough capacity to meet the flow specification

22

• Choke Packets (for both VC and datagram)
• by monitoring the utilization of a line a choke
  packet can be sent to the sender to reduce the
  transmission by certain
• after some period after the choke packet the
  sender can come back to the previous rate
• Weighted Fair Queueing
• if n hosts are sharing an output line then n
  internal queues can be maintained as a round
  robin fashion
• some ATM switches use this method
• byte by byte or packet by packet round robin
• file servers can be given a higher priority
• give more bytes per tick
• Hop-by-Hop Choke Packets
• for a long distance the choke packet takes some
  time to get to the source
• for a high speed this period can generate a lot
  of traffic
• (a) regular choke packet
• (b) Hop-by-Hop choke packet

23

• Load Shedding
• when none of the above works then router will
  drop packets
• but which one?
• For a file transfer
• dropping 6 out of 12 may require retransmitting 6
  to 12
• dropping 10 may require retransmitting only 10 to
  12
• some packets are more important than the others
  depending on the application
• a line of pixels in images is less important than
  a line of text
• packet types can be priotized
• ATM has 1 bit in header allocated for priority
• Jitter Control
• audio and video applications delay is less
  important than the variation of the delay
• as long as frames arriving in reasonable delay
  variations its acceptable
• jitter can be controlled at the router by trying
  to keep the space between the packets constant
• Congestion Control for Multicasting (Resource
  reSerVation Protocol)
• multicast spanning tree is built for each senders
• using these spanning trees each receiver makes
  bandwidth reservation request up to the sender

24
InterNetworking

• In reality, we dont have a single homogeneous
  network
• we have something looks like this

• For Internetworking we need at each layer
• repeaters that just copy bits between cable
  segments (layer 1)
• bridges store and forward data link frames
  between LANs (layer 2)
• multiprotocol routers forward packets between
  dissimilar networks (layer 3)
• transport gateways connect byte streams in the
  transport layer
• application gateways allow interworking (above
  layer 4)
• (b) gateway between LAN and WAN
• (c ) half gateways

25

• How Networks Differ
• the table explains all

26

• Concatenated Virtual Circuits
• a sequence of virtual circuits is set up from the
  source through one or more gateways to the
  destination
• each gateway relays incoming packets by
  converting between packet formats and virtual
  circuit numbers, etc.
• each gateway has a table indicating incoming and
  outgoing VC number mapping
• it can be full or half gateways
• ex) OSI -gt TCP -gt OSI
• the concatenation can change the nature of
  service
• if source network provides reliable delivery of
  network layer packets
• but if the via network doesnt
• then it is no longer an option

27

• Connectionless Internetworking
• if each network has its own network layer
  protocol, it is not possible for a packet from
  one network to transit to another
• unless one format is very close to another, the
  conversion is difficult and rarely attempted
• addressing is difficult
• IP address is 32 bit
• mapping one protocols address into another is
  not simple
• to be fully compatible each host must have
• OSI address
• IP address
• xyz address
• Guess how many hosts in the world!
• How can you assign them?
• How can Multiprotocol remember/record these in
  order to translate?

28

• Tunneling
• internetworking is tolerable if the source and
  destination networks are of the same type
• Tunneling can be used in between
• ethernet LAN to PTSNs point to point WAN to
  ethernet LAN
• this job is manageable by Multiprotocol router

29

• Internetwork Routing
• An internetwork with multiprotocol gateways can
  be made into a graph and then
• a two level routing algorithm can be used
• an exterior gateway protocol between the networks
• a packet is sent to the local multiprotocol
  router
• the network layer sw decides which multiprotocl
  router to forward this packet to using its own
  multiprotocol router routing table
• it may have to be tunneled if via network is
  different
• an interior gateway protocol within each network
• each network is called an Autonomous System since
  they are independent to each other
• it may well be (more often than not) over
  international boundary line - more trouble
• each country has her own rules and regulations
• Billing is a big issue since each network has
  different charge plan

30

• Fragmentation
• reasons to have some maximum size on the packets
• HW (the width of a TDM transmission slot)
• Operating System (buffer size of OS is 512 bytes)
• Protocols (the number of bits in the packet
  length field)
• (inter)national standards
• attempt to reduce error induced retransmissions
  to some level
• attempt to prevent one packet from occupying the
  channel too long
• different packet sizes for different networks
  introduce packet fragmentation
• (a) transparent packet fragmentation
• suit yourself to the size of the via network
• the next network doesnt even know fragmentation
  happened in the previous network
• (b) nontransparent fragmentation

31

• Fragmentation - continues
• (a) transparent packet fragmentation
• the exit gateway must know when and whether it
  received all the pieces
• all packets fragments must exit via the same
  gateway
• (b) nontransparent fragmentation
• the destination host does the reassemble
• it requires all the hosts to be able to assemble
• fragmentation overhead remains for the whole trip
  to destination
• tree structure can be used in constructing
  fragmentation id
• packet 0 can be fragmented into 0.1, 0.2, and 0.3
  and
• further fragmentation can be 0.0.0, 0.0.1, 0.0.2,
  etc.
• reassemble problem
• if 0.0 gets lost and the original packet is
  retransmitted
• if the original packet goes through smaller
  packet sized network and it becomes 0.0, 0.1,
  etc. but twice as many fragmentation
• the same 0.0 is arrived but half the size
• its tough for the assembling gateway
• original packet number and the fragmentation
  number
• can be used
• (a) original packet w/ 10 bytes

32

• Firewalls
• to protect from hackers, viruses, industry
  espionage, corporations surround their networks
  with firewalls
• all communication go through some kind of filter
  (guess why internet access is slow?)
• packet filtering router
• packet by packet inspection is done
• port abling and disabling
• port 23 for telnet, port 79 for finger, etc.
• sometimes ports are assigned dynamically, tough
• application gateway
• look at the mail header, etc.
• hard to filter wireless and radio communications

33
The Network Layer in the Internet

• The Internet is an interconnected collection of
  many networks
• The Internet is supported by its network layer
  protocol called IP (Internet Protocol)
• designed from beginning with internetworking in
  mind
• for better explanation of IP please see Douglas
  E. Comer, Internetworking with TCP/IP, vol 1, PH

34

• The IP Protocol
• The IP header
• 20 byte fixed part (first 5 rows)
• variable length optional part
• transmitted in big endian order from left to
  right, high order bit to low order bit
• Version IP version number
• HLEN or IHL header length in 32 bit words
  (remember the variable optional part)
• minimum is 5 with no options
• maximum is 15 (24-1) or 15 4 60 bytes for
  header(40 byte or 10 rows max for option part)
• Type of Service 8 bit (Precedence(3),D,T,R bits,
  unused(2))
• Precedence priority (0..7network control
  packet)
• D low delay preferred
• T high Throughput preferred
• R high Reliability preferred
• it may not be possible for the Internet to
  guarantee the type of transport requested
• consider it as a hint to the routing algorithm
  rather than a demand
• Total Length includes data (216 65,535 bytes)
• header is upto and include options
• data follows immediately after options

35

• The IP header - continues
• Identification packet ID useful when reassemble
  fragments
• Flags
• DF Dont Fragment sender knows it may have to
  avoid smaller packet size networks
• for certain control packets where only the whole
  length is useful
• if a router cant avoid fragmenting, this packet
  will be discarded and an error msg is sent back
• MF More Fragments will come the last fragment
  doesnt have this bit set
• Fragment Offset offset of this fragment in the
  original datagram/packet
• all fragments are multiple of 8 bytes except the
  last one
• first fragment will have this value as 0
• Once the datagram with MF bit turned off
• the router or destination node whoever does the
  reassembly can tell whether all the fragments
  have been received
• by looking at Total Length and Fragment offset
  fields of each fragment datagrams received
• Time to Live to limit how long, in seconds, the
  datagram is allowed to remain in the internet
  system
• since its hard to measure the transit time for
  physical networks, it is decremented at each hop
• plus a router records the arrival time and
  decrements the duration the datagram remained in
  the router when leaving
• when 0, this datagram is discarded and an error
  message is sent to the sender
• Protocol high-level protocol that was used to
  create the message being carried in the data area
  of datagram (TCP or UDP for example)

36

• The IP header - continues
• Header Checksum ensures the integrity of header
  values
• Source address 32 bit
• Destination address 32 bit
• Options
• to allow information that are not part of header
  for later versions of protocols
• some information is rarely used and not worth
  being part of regular header
• padded to make a multiple of 4 bytes
• not all options are supported by all routers
• Option Octet copy (1 bit), option class (2
  bits), option number (5 bits)
• copy 1 the option should be copied into all
  fragments
• option class
• 0 datagram or network control
• 1 reserved for future use
• 2 debugging and measurement
• 3 reserved for future use
• option number class, number, length
• 0,0,- end of option list
• 0,1,- no op

37

• The IP header - continues
• Security option for mulitary do not route this
  packet through Cuba
• if its really secret, dont use internet use
  James
• Record Route option format
• option octet, length octet (total length of the
  option including the first 3 octets), pointer
  (offset of the next available slot), only 3
  octets no padding to make it 4
• first IP address
• second IP address
• they provide a way to monitor or control how
  routers route datagrams
• enough space must be allocated in the option by
  the original source to hold all entries that will
  be needed
• if full (the pointer gt the length) then no more
  address is inserted
• Strict Source Routing (the same format as Record
  Route option)
• source node dictates the route
• useful when routing tables are corrupted or for
  timing measurements
• an error occurs if the exact path can not be
  followed
• Loose Source Routing
• source node dictate only few important routes
• allows other routes in between specified
  routes/routers
• for both Source Routing methods, if the list is
  exhausted before reaching the destination, the
  remaining segments is routed normally as if no
  source routing

38

• IP Addresses
• unique 32 bit address for every host in the
  Internet - the Internet is running out of address
  space (the number of hosts doubles each year) -gt
  IPv6 with bigger address space, 128 bit
• each address is a pair (netid, hostid)
• class A bigger networks (1,7,24 bits) of
  hosts gt 216
• 9.0.0.0 IBM 12.0.0.0 ATT
• class B intermediate networks (2,14,16 bits) 28
  lt of hosts lt 216
• class C smaller networks (3,21,8 bits) of
  hosts lt 28
• network ids are assigned by the INTERNIC
  (Internet Network Information Center)
• multi-homed hosts hosts that are connected to
  more than 1 network requires multiple IP addr
• IP addresses encode both a network and a host,
  they do not specify an individual computer, but
  rather a connection to a network
• if a host moves from one network to another, its
  IP address must change
• multicast address the address refers to a
  multicast group
• Loopback address 127 is for the class A range
• to test TCP/IP and for inter-process
  communication on the local machine

39

• Subnets
• different from Subnet in network (WAN)
• How can one minimize the number of assigned
  network addresses, especially class B, without
  destroying the original addressing scheme?
• Allow a network to be split into several parts
  for internal use but still act like a single
  network to the outside world These parts are
  called subnets
• outside the network, this subnetting is not
  visible
• Once a packet arrives, the Subnet mask is ANDed
  with the address to find the subnet number
• Each Subnets router needs to remember only the
  hosts on the subnet reducing the routing table
  entries

40

• Internet Control Protocols
• Internet Control Message Protocols (ICMP)
• routers closely monitor the Internet and
  unexpected events are reported to ICMP to test
  the Internet
• It is an error reporting mechanism
• when a datagram causes an error, ICMP can only
  report the error condition back to the original
  source
• the source of the datagram must relate the error
  to an individual application program or take
  other action to correct the problem (like
  reporting an alarm to the administrator)
• why to the source only?
• Datagram finds its way (route) at each step of
  the way has source and destination address
  only
• cant tell if a router in the middle has a
  routing table problem
• must rely on the host administrator and the
  network administrators to locate and repair the
  problem
• It provides communication among routers and
  hosts
• It is an integral and required part of IP

41

• The Address Resolution Protocols (ARP)
• each host has an assigned IP address and a
  physical address (board address)
• Es are ethernet addresses
• Fs are FDDI addresses
• the Data Link layer doesnt understand IP
  addresses
• IP addresses need to get mapped onto data link
  layer addresses such as Ethernet (48 bit board
  address)
• The problem of mapping high-level addresses to
  physical addresses is known as the address
  resolution problem
• resolution through direct mapping put IP and
  Ethernet addresses in system configuration file
• hard to maintain and error will be introduced by
  the wrong/old information
• resolution through dynamic binding ARP
• send broadcast message asking who has this IP
  address?
• The owner will respond with the ethernet address
• IP packet is encapsulated (source and destination
  IP addresses) into the ethernet packet (source
  and destination ethernet addresses) and sent
• both source and destination hosts can keep this
  information in cache for immediate future use
• ARP is a low-level protocol that hides the
  underlying network physical addressing,
  permitting one to assign an arbitrary IP address
  to every machine. Think of ARP as part of
  physical network system, and not as part of the
  internet protocols
• a router can deliver ARP messages to other LANs
  called proxy ARP

42

• The Reverse Address Resolution Protocols (RARP)
• reverse problem to ARP
• Given an ethernet address, what is the IP
  address?
• For a diskless system like some SUN workstations,
  their IP address information is kept on a remote
  server
• Upon booting up, hosts sends a broadcast message
  asking (This is my ethernet address. Who has my
  IP address?) using RARP
• Since RARP is using all 1s to reach the RARP
  server, RARP server must be on each LAN
• limited to its own LAN
• BOOTP (BOOTstrap Protocol) a more recent
  version
• RARP uses low level address determination
• BOOTP and DHCP (Dynamic Host Configuration
  Protocol) build on higher level protocols like IP
  and UDP to go over routers

43

• The Interior Gateway Routing Protocols (IGP)
  OSPF,RIP,HELLO
• IGP Interior Group Protocols
• AS (Autonomous System) the Internet is made of
  a large number of As (various sized networks)
• routing protocols used within AS are called IGP
• RIP Routing Information Protocol the Vector
  Distance Routing
• HELLO protocol now obsolete the Vector
  Distance Routing but uses network delay iso hop
  counts
• OSPF Open Shortest Path First
• OSPF
• RIP was the original Internet interior gateway
  protocol which didnt work well with larger ASs
• remember the count to infinity problem?
• OSPF is the successor that became a standard in
  1990
• Open, supports various physical distance and
  delay, etc., dynamic algorithm, supports type of
  service, load balancing by splitting the load
  over multiple lines, supports hierarchical
  systems (the Internet getting too big, no router
  is expected to know the entire topology),
  security, provisioning to deal with routers
  connected to the Internet via a tunnel
• AS can be represented as a weighted graph

44

• The Interior Gateway Routing Protocols (IGP)
  OSPF - continues
• ASs can be divided into areas
• within an area, each router has the same link
  state database and runs the same shortest path
  algorithm
• intraarea routing
• interarea routing
• interAS routing
• internal routers within one area
• area border routers between two or more areas
• backbone routers on the backbone
• AS boundary routers to talk to other ASs
• these 4 routers can overlap
• type of service is handles by creating separate
  graphs using the delay, throughput, and
  reliability
• the table shows the types of OSPF messages

45

• The Exterior Gateway Routing Protocol (EGP) BGP
• BGP (Border Gateway Protocol) is used between
  ASs
• BGP must consider politics
• no transit through Cuba for Pentagon packets
• no transit outside Canada for Canadian packets
• no transit through MS for IBM packets
• Politics are manually configured into each BGP
  routers and not part of the protocol itself
• networks (ASs) are classified
• stub networks like a leaf of a tree cant be
  used as a transit network
• multiconnected networks non stub networks but
  they refuse to be used as a transit network
• transit network willing to be used as a transit
  but with some restrictions
• fundamentally a Distance Vector Protocol but
  different from others like RIP
• maintains more than just the cost to each
  destination
• keeps track of the exact path used
• each node periodically sends its neighbors the
  exact path it is using
• the receiving node recalculates the best route to
  a destination
• A .. J routers representing networks (ASs)
• (a) a set of BGP routers
• (b) information sent to F router

46

• Internet Multicasting
• IP supports multicasting using type D addresses
• each class D identifies a group of hosts
• permanent group address
• 224.0.0.1 all systems on a LAN
• 224.0.0.2 all routers on a LAN
• 224.0.0.5 all OSPF routers on a LAN
• 224.0.0.6 all designated OSPF routers on a LAN
• temporary group address
• a best effort is used and some members of a group
  may not get the packet
• IGMP (Internet Group Management Protocol) is
  used
• close to ICMP
• only query and response packets
• Multicast routing is done using spanning trees

47

• Mobile IP
• traveling IP user
• see Routing for Mobile Users
• this user has a home agent in NYC and has a fixed
  IP address
• this is not usable in LA
• the mobile user registers to LAs foreign agent
  and start using its guest IP address
• 0 LAs foreign agent sends this guest IP address
  to the home agent in NYC
• 1 a sender sends data to the mobile users NYC
  home IP address
• 2 NYC home agent sends this data with the
  senders IP address to LAs foreign agent
• 3 NYC home agent sends the mobile users guest
  IP address to the sender in Seattle so that the
  future communication can be done right without
  bothering NYCs home agent
• 4 the sender establishes tunneling to the mobile
  users guest IP address in LA

48

• CIDR - Classless InterDomain Routing
• IP address space is running out quickly Running
  Out of Address Space (ROADS) problem
• class A millions of hosts (class ID1,net
  ID7,host ID24)
• class B thousands of hosts (2,14,16)
• class C upto 254 (3,21,8)
• three bears problem
• most organizations think class A network is too
  much
• most organizations think class C network is too
  small
• most organizations think class B network is just
  right popular demand
• in reality
• there are many class B networks but
• class B holding organizations have fewer than 50
  hosts
• it might have been better if class C used 10 bits
  instead of 8 bit host Ids
• to conserve class B addresses, many class C
  addresses were given out instead
• created problem with routers
• routers have to store much more information
• a technique known as Classless Inter-Domain
  Routing (CIDR) solves the problem temporarily
• CIDR
• a block of contiguous class C addresses are
  collapsed into a single entry represented by a
  pair

49

• CIDR - Classless InterDomain Routing - continues
• suppose an organization was assigned a block of
  2048 contiguous addresses starting at address
  234.170.168.0 (class D for multicast routing is
  used!)
• lowest 234.170.168.0 11101010 10101010 10101000
  00000000
• highest 234.170.175.255 11101010 10101010
  10101111 11111111
• CIDR requires 2 values to specify the range of
  values the lowest address and 32 bit mask
• 11111111 11111111 11111000 00000000
• see 11 0s to give 211 2048 when masked
• with the base (lowest) address and the mask, the
  address to this organization can be figured out
• routing software does not interpret the
  destination address class
• instead, each entry in the routing table contains
  an address and a mask
• called Supernet Addressing or Supernetting as
  opposed to subnet addressing
• since it allows the use of many IP network
  addresses for a single organization
• current IPv4 uses 32 bit addressing where the
  next IPv6 uses 128 bit addressing
• IPv5 in use for an experimental real-time stream
  protocol

50

• IPv6
• address space was/is running out
• IPv4 needed to be made more flexible with many
  service introductions
• Deering and Francis proposals in 1993 were
  selected and called SIPP (Simple Internet
  Protocol Plus) later given IPv6
• not fully compatible with IPv4 but reasonably
  well with others TCP,UDP,ICMP,IGMP,OSPF,BGP, and
  DNS
• longer address space, 128 bit
• simplification of the header 7 fields vs 13 in
  IPv4 packets are routed faster
• better support for options many previously
  required fields are now made optional
• much better security Authentication and
  Security are key features in IPv6
• type of service is made better (8 bit field) but
  may not be enough
• the transition will take take a decade
• The Main IPv6 Header
• version field is always 6
• routers will have to examine this field to tell
  which packet they have
• wasteful
• many will create a field in the data link header
  to distinguish v4 and v6
• and have 2 network layer handlers
• violation of layered approach forcing the data
  link layer to know the network layer packet
  types

51

• The Main IPv6 Header
• The IPv6 fixed header (required)
• priority field to distinguish flow controllable
  sources from non flow controllable sources
• 0..7 in terms of slowing down capability during
  congestion
• 8..15 for real time traffic with constant flow
  audio and video dont like jitters (delay
  variations)
• allows routers to handle different types of
  packets better during congestion
• Flow Label
• to allow a source and destination to set up a
  pseudoconnection with particular properties and
  requirements
• like delay requirement and thus may need to
  reserve bandwidth
• flow can be setup in advance and an ID can be
  given
• routers look up the table to see what kind of
  special treatment this ID requires
• like a combination of the flexibility of a
  datagram subnet and the guarantees of a virtual
  circuit subnet
• flow numbers are to be randomly generated for
  easy hashing

• Payload length
• how many bytes follow the 40-byte header
• different from Total Length in IPv4 since it
  doesnt include the header length
• Next Header
• there are 6 extension headers
• this field tells which of the 6 types follow this
  header if any
• if this header is the last IP header, this field
  tells which transport protocol handler (TCP,UDP)
  to pass the packet to
• Hop limit
• the same as the Time to Live field in IPv4 (in
  seconds)
• Source and Destination address
• 16 bytes or 128 bits

52

• The Main IPv6 Header - continues
• address starts with different prefixes (128 bit
  16 bytes)
• Ipv4 addresses prefixed by 80 zeros (10 byte
  zeros)
• next 16 bytes determines two variants
• these variants tell how IPv6 packets will be
  tunneled over the IPv4 infrastructure
• Provider based (010 prefix) ATT, MCI, Sprint,
  BT, etc.
• next 5 bits indicate one of the North America,