COM 360

1
COM 360
2
Chapter 4

• Internetworking

3
Interconnecting Networks

• Problem Theres more than one network
• Problems of Heterogeneity and Scale
• Heterogeneity- users on one type of network want
  to be able to communicate with users on other
  types of networks.
• Internet Protocol(IP) and how it is used to build
  heterogeneous, scalable networks.
• Principle of Routing- finding loop-free paths
• The problem of the growth of the Internet- going
  from IPv4 to IPv6

4
Simple Internetworking

• What is an internetwork?
• Internetworks or internets (with lower case i)
  are large, highly heterogeneous networks with
  reasonably efficient routing.
• They are a collection of networks that are
  interconnected to provide host-to-host packet
  delivery service.
• With a capital I the Internet refers to the
  global Internetwork.

5
What is an Internetwork?

• What is the difference between networks, subnets
  and internets?
• A network is a directly connected or switched
  network, which uses a single technology (802.5,
  Ethernet, or ATM) and represents a physical
  network.
• A subnet uses single IP address to denote
  multiple physical addresses.
• An internet is a collection of networks or
  logical networks, built out of a collection of
  physical networks.

6
A Simple Internetwork

• An internetwork is referred to as a network of
  networks because it is made up of many smaller
  networks.
• For example, an internetwork can connect
  Ethernets, FDDI rings and Point-to-links (See
  next slide)
• The nodes that connect them are called routers
  (and sometimes gateways)
• The Internet Protocol is the tool used to build
  heterogenous internetworks.

7
A Simple Internetwork
HN host Rn router
8
Internet Protocol (IP)

• IP is the tool used to build scalable,
  heterogeneous internetworks.
• Originally called the Kahn-Cerf protocol after
  its inventors.
• IP runs on all the hosts and routers and defines
  the infrastructure that allows them to function
  as a single network.

9
A Simple Internetwork
Shows Protocol layers used to connect H1 to
H8 ETH is the Protocol that runs over the Ethernet
10
Service Model

• When you build an internet, start by defining the
  service model, or the host-to-host services that
  you want to provide, over each of the underlying
  physical networks.
• An addressing scheme, which provides a way to
  identify all hosts in the internet
• A datagram (connectionless) model of data
  delivery.
• This service model is called best effort, because
  although IP makes every effort to deliver
  datagrams, it makes no guarantees.

11
Datagram Delivery

• IP datagram is fundamental to the Internet
  Protocol
• A datagram is a type of packet that is sent in a
  connectionless manner over a network.
• Every datagram carries enough information to let
  the network forward the packet to its
  destination.
• No set up mechanism is needed just send it and
  the network tries to get it to its destination.

12
Best Effort Delivery

• If something goes wrong and a packet gets lost,
  corrupted or misdelivered, or in any way fails to
  reach its destination, the network does nothing.
  It is called unreliable service.
• Best-effort, connectionless service is the
  simplest service for an internetwork.
• Keeping the routers as simple as possible was one
  of the original design goals of IP.
• The ability of IP to run over anything is its
  most important characteristic.

13
Data Transmission and Frames

•    IP internet layer
•  Constructs datagram
•  Determines next hop
•  Hands to network interface layer
• Network interface layer
•  Binds next hop address to hardware address
•  Prepares datagram for transmission
•  But ... hardware frame doesn't understand IP
  how is datagram transmitted?

14
Datagram Encapsulation

• Network interface layer encapsulates IP datagram
  as data area in hardware frame
•  Hardware ignores IP datagram format
•  Standards for encapsulation describe details
•  Standard defines data type for IP datagram, as
  well as others (e.g., ARP)
•  Receiving protocol stack interprets data area
  based on frame type

15
Encapsulation in a Hardware Frame
16
Transmission Across an Internet

• Each router in the path from the source to
  the destination
•  Unencapsulates incoming datagram from frame
•  Processes datagram - determines next hop
• Encapsulates datagram in outgoing frame
•   Datagram may be encapsulated in different
  hardware format at each hop
• Datagram itself is (almost!) unchanged

17
Datagram Transmission
18
Datagram Delivery

• A datagram is fundamental to the IP portocol
• A datagram is sent in a connectionless manner
  over a network
• Best effort if something goes wrong, the
  network does nothing.
• Simples type of service- keeping routers simple
  was one of the design goals
• Ability of IP to run over anything- main
  advantage (even a network of carrier pigeons!??)

19
IP Packets

• Part of the IP service model is the type of
  packets it can carry.
• IP datagram consists of a header followed by the
  number of bytes of data.
• These are usually represented by 32-bit words,
  where the top word and the leftmost words are
  transmitted first.

20
IPv4 Packet Header
21
Packet Format

• There are some subtle features of this simple
  model
• The Version field specifies the current IP
  version, called IPv4. Putting it first makes it
  easy to define everything else.
• HLEN specifies the length of the header (about 5
  words or 20 bytes).
• TOS- is the Type of Service field
• The LENGTH field (in bytes)- length of datagram,
  including the header

22
Packet Format

• TTL- is the Time to Live field, set to a specific
  number (64 is current default) which the routers
  would then decrement, until it reached 0. It
  purpose is to discard packets that have been
  circling around and to discard them.
• Protocol field identifies the higher level
  protocol (TCP, UDP) to which this packet should
  be passed.
• Checksum- add the entire header and take the ones
  complement of the result.

23
Packet Format

• SourceAddr- Source Address enables a recipient
  to reply
• DestinationAddr - Destination Address this is
  key to the delivery of the datagram
• IP defines its own global address space,
  independent of the physical network
• There are also optional fields, which are rarely
  used.

24
Fragmentation and Reassembly

• Each network technology has its own maximum
  packet size
• (Ethernet 1500 bytes, FDDI- 4500 bytes)
• Two choices for the IP service model
• Make sure all IP datagams are small enough or
• Provide a means by which packets can be
  fragmented and reassembled, when they are too big
  to be sent though a network technology

25
Fragmentation and Reassembly

• Every network has a maximum transmission unit
  (MTU), which is the largest IP datagram that it
  can carry in a frame.
• This value is smaller than the largest network
  packet size, because it must fit into the payload
  of the data link layer frame.
• When a host sends a datagram it can choose any
  size. A reasonable choice is the MTU of the
  network to which it is directly attached.

26
Datagram Fragmentation

• Router uses local MTU to compute the size of each
  fragment and puts part of the original data in
  each fragment and rest of the information in the
  header.

27
Fragmentation and Reassembly

• Fragmentation will be necessary if the path to
  the destination includes a destination with a
  smaller MTU.
• Fragmentation typically occurs in a router (in
  IPv4).
• To enable the fragments to be reassembled at the
  receiver, each datagram carries the same
  identifier in the ident field.
• The unique identifier is chosen by the sender.
• If all fragments do not arrive at the receiver,
  it discards all datagram fragments and does not
  attempt to recover them.

28
Fragment Loss

•   IP may drop fragment
•   What happens to original datagram?
•    Destination drops entire original datagram
•    How does destination identify lost fragment?
•    Sets timer with each fragment
•    If timer expires before all fragments arrive,
  fragment assumed lost
•    Datagram dropped
•   Source (application layer protocol) assumed to
  retransmit

29
IP Datagram Traversing the Sequence of Physical
Networks
This is what happens when H1 sends a datagram to
H8. Assume 1500 bytes for an Ethernet, 4500 for
FDDI, 532 for PPP. The datagram is broken into 3
fragments at router 2, which are then forwarded.
30
Fragments

• Each fragment is a self-contained IP datagram
  that is transmitted over physical networks,
  independent of the other fragments.
• Each IP datagram is re-encapsulated for each
  physical network over which it travels.
• Fragmentation is done in 8 byte chunks.
• The router sets the M bit in the FLAGS field to
  indicate there are more fragments, and sets the
  OFFSET field to zero to indicate the first part
  of the datagram.

31
Header Fields Used in IP Fragmentation
a) Unfragmented packet
b) Fragmented packets
32
Fragmentation

• Fragmentation is done in such a way that it could
  be repeated if a fragment arrived at another
  network with an even smaller MTU.
• The fragments are easily reassembled independent
  of the order in which they are received.
• Reassembly is done at the receiving host and not
  at each router. Why?
• (See p. 243-247 for reassembly code.)

33
Path MTU Discovery

• IP reassembly is not a simple process and should
  be avoided. ( For example, if a fragment is lost,
  the receiver still tries to reassemble the whole
  datatgram until it finally must discard it.)
• Instead, hosts are encouraged to perform path
  MTU discovery by sending packets small enough to
  go through the path with the smallest MTU form
  sender to receiver. It first sends large
  datagrams, and if they are not successful, then
  is sends smaller ones, until it discover the
  smallest MTU from sender to receiver.

34
Global Addresses

• Global uniqueness is the first property of an
  addressing scheme.
• Ethernet addresses are flat and without
  structure.
• IP addresses are hierarchical and are made up of
  several parts that correspond to parts of the
  network.
• IP addresses consist of a network part and a host
  part.

35
Addresses For the Internet

•   One difference between an internet and a
  physical network is that an internet is an
  abstraction imagined by its designers and created
  by software.
• Designers choose addresses, packet formats, and
  delivery techniques independent of the hardware.
•   One key aspect of virtual network is single,
  uniform address format
•   Each address must be unique
•   Can't use hardware addresses because different
  technologies have different address formats.

36
IP Addressing Scheme

•   Addressing in TCP/IP is specified by the
  Internet Protocol (IP)
•   Each host is assigned a 32-bit number
• (4 octets, separated by dots) referred to as
  dotted octet ( e.g. 216.72.32.10)
•   Called the IP address or Internet address
•   Unique across entire Internet
• Different from a domain name linux.sjcny.edu

37
IP Address Hierarchy

•   Each IP address is divided into a prefix and a
  suffix
•   Prefix identifies network to which computer is
  attached
•   Suffix identifies computer within that network
  
•   Each physical network is assigned a unique
  network number
•   Address format makes routing efficient
•  Each computer is assigned a unique address
•  Network assignments are coordinated globally
  but suffixes can be assigned locally.

38
IP Addresses

• The network part of the IP address identifies the
  network to which the host is attached
• All hosts attached to the same network have the
  same network part in their IP address.
• The host part or suffix, identifies each host
  uniquely on that network.

39
Classes of IP Addresses

•   Designers chose a compromise - multiple
  address formats that allow both large and small
  prefixes
•   Original scheme called classful IP addressing,
  divided the IP address space into 3 primary
  classes, where each class had a different size
  prefix and suffix
•   Each format is called an address class
• Class of an address is identified by first four
  bits

40
IP Addressing

•   Octet (8-bit) boundaries are used to
  partition an address into prefix and suffix
•   Class A, B and C are primary classes
•   Used for ordinary host addressing
•   Class D is used for multicast, a limited form
  of broadcast
•   Internet hosts join a multicast group
•   Packets are delivered to all members of group
•   Routers manage delivery of single packet from
  source to all members of multicast group
•   Used for MBone (multicast backbone)
•   Class E is reserved ( for future use)

41
Computing the Class of an Address

•   IP software computes the class of the
  destination address when it receives a packet.
•   IP addresses are self-identifying because the
  class can be computed directly from the first few
  bits of the address
•  The first 4 (leading) bits of the address
  denote the class
• Class A begins with 0
• Class B begins with 10
• Class C begins with 110

42
Computing the Class of an Address
43
IP Address Classes
Prefix designates the network, suffix designates
the host.
44
Primary IP Address Classes
a) Class A
B) Class B
C) Class C
45
Dotted Decimal Notation

•  Class A, B and C all break between prefix and
  suffix on byte boundary
•  Dotted decimal notation is a convention for
  representing 32-bit internet addresses in decimal
  
•  Convert each byte of address into decimal
  separate octet by periods ("dots'')
•  Dotted decimal notation treats each octet as an
  unsigned binary integer
•  Smallest value is 0.0.0.0 and largest is
  255.255.255.255

46
Dotted Decimal Notation
What would SJCs address be in binary (
216.73.32.0)?
47
Classes and Dotted Decimal Notation

•  While dotted decimal makes separating network
  address from host address easier, determining
  class is not so obvious
•  Look at first dotted decimal number, and use
  this table to calculate the class

48
Division of Address Space

• Addressing scheme is flexible and allows networks
  of various sizes to be accommodated efficiently
• Original idea was that Internet would consist of
  small number of wide area networks (Class A), a
  few site (or campus) sized (Class B) networks,
  and a large number of LANs (Class C)
• Additional flexibility was needed and removed
  some of the distinction between classes present
  in this classful scheme.

49
Division of Address Space

•  IP Class scheme does not yield equal number of
  networks in each class
•  Class A
•  First bit must be 0
•  7 remaining bits identify Class A net
•  27 ( 128) possible class A nets
•  Number of bits allocated to a prefix or suffix
  determines how many unique numbers can be
  assigned
•  A prefix of n bits allows 2n unique network
  numbers, while a suffix of n bits allows 2n hosts
  number on a given network

50
Number of Networks and Hosts
51
Authority for Addresses

• An organization obtains unique network numbers
  from an Internet Service Provider (ISP), which
  coordinates with the Internet Assigned Number
  Authority. A network administrator can assign
  prefixes in a private internet.
• (See Internic, ICANN, Educause, etc.)

52
Insufficient Addresses

•  Large organizations may not be able to get as
  many addresses in the Internet as they need
•  Example - UPS needs addresses for millions of
  computers
•  One solution - set up private internet and
  allocate addresses from entire 32-bit address
  space
• Others do not use all their assigned addresses
• For example, SUNY Stony Brook has a Class B
  license but probably only uses 3000-40,000
• of its 216 addresses (65,536 possible).

53
A Classful Addressing Scheme

•   Select address class (usually class C) for
  each network depending on expected number of
  hosts
•   Chosen by the internet service provider for
  the internet
•   Chosen by the network administrator in a
  private network
•   Assign network numbers from appropriate
  classes
•   Assign host suffixes to form internet
  addresses for all hosts

54
Classful Addressing with Private internets

• Consider an organization with a private TCP/IP
  with 4 networks, connected by routers.
•  A prefix is chosen denoting the class (A,B,C)
  depending on the size of the network
•  In the next example, there is one Class A
  network
• (prefix 10), two class B prefixes (128.100 and
  128.211) and one class C (192.5.48).
•  The IP address assigned to the host begins with
  the prefix assigned to the host's physical
  network
•  Suffixes, which are assigned by the local
  network administrator, can be arbitrary numbers,
  often chosen sequentially.

55
Classful Addressing with Private internets
56
Subnet and Classless Addressing

• Two mechanism were invented to overcome the
  addressing limitations
• 1.     Subnet addressing
• 2.     Classless addressing
• These are so closely related that they can be
  thought of as a single abstraction instead of
  having 3 distinct address classes, allow the
  division between prefix and suffix to occur on an
  arbitrary bit boundary.

57
Address Masks

• How can an IP address be divided at an arbitrary
  boundary?
• It requires an additional piece of information to
  be stored with each address. This information
  specifies the exact boundary between the network
  prefix and the host suffix.

58
Address Masks

• To use classless or subnet addressing the routers
  must store 2 pieces of information
• the 32 bit address and
• another 32 bit value that specifies the boundary
  between the prefix and suffix.
• This second value is called the called the subnet
  mask and 1 bits mark the network prefix and zero
  bits mark the host portion. This makes
  computation efficient.

59
Routers and Addresses

• Routers compare the network prefix portion of the
  address to a value in their routing tables.
• Suppose a router is given a destination address,
  D and a pair (A,M) that represents the 32 bit
  address and the 32 bit subnet mask.
• To make the comparison, the router tests the
  logical "and" condition to set the host bits of
  address D to zero and then compares the result
  with the network prefix A
• A ( D M)

60
Routers and Addresses

• For example consider this 32 bit mask
• (255.255.0.0 in decimal)
• 11111111 11111111 00000000 00000000
•   and the network prefix (128.10.0.0 in decimal)
• 10000000 00001010 00000000 00000000
• Now consider the 32 bit destination address
  128.10.2.3 which has the binary equivalent of
• 10000000 00001010 00000010 00000011
• The logical "and" between the destination address
  and the address mask produces the result
• 10000000 00001010 00000000 00000000
• which is equal to the prefix 128.10.0.0

61
CIDR (Classless Interdomain Routing) Notation

• Inside the computer each address mask is stored
  as a 32 bit value in binary, which is then
  expressed in dotted octet notation.
• The new CIDR notation append a slash and the size
  of the mask in decimal notation
• For example 128.10.0.0/16

62
CIDR Address Block Example

• Suppose an ISP has a single Class B license
  128.211.00.0. Using a classful address scheme,
  he/she can only assign the prefix to one
  customer, who can have up to 216 host addresses.
• Using CIDR, the ISP could assign the entire
  prefix to a single organization by using
  128.211.0.0/16
• Or he could partition the address into three
  pieces (two of them big enough for 2 customers
  with 12 computers each and the remainder
  available for future use.

63
CIDR Address Block Example

• One customer could be assigned 128.211.0.16/28
• and the other could be assigned 128.211.0.32/28
• Both customers have the same mask size (28 bits),
  but the prefixes differ and each has a unique
  prefix. More importantly the ISP retains most of
  the addresses, which can then be assigned to
  other customers.

64
CIDR Host Address
65
Special IP Addresses

• IP assigned a set of addresses that are reserved
  and never assigned to hosts
• Network Address
• IP reserves host address zero and uses it to
  denote a network. (For example,128.211.0.0 is a
  Class B network)
• Direct Broadcast Address
• It is formed by adding a suffix consisting of all
  1's to the network prefix (For example,
  128.211.111.111)
• Limited Broadcast Address
• A broadcast on a local physical network (or
  limited to a "single wire") is used during system
  startup by a computer that does not yet know the
  network number. The address with all 1's is a
  limited broadcast.

66
Special IP Addresses

• This Computer Address
• A computer needs to know its own IP address to
  send or receive internet packets. The TCP/IP
  protocol allows a computer to obtain its address
  automatically but strangely enough, when using
  these startup protocols the computer cannot
  supply a correct IP source address. To handle
  such cases, IP reserves the address that consists
  of all zeroes to mean "this computer".

67
Special IP Addresses

• Loopback Address
• A loopback address is used to test network
  applications. IP reserves the network prefix 127
  for use with loopback and programmers usually use
  the host number 1 (forming the address 127.0.0.1)
  for loopback testing.
• During loopback no packets actually leave the
  machine - the IP software forwards packets from
  one application program to another on the same
  computer. Therefore the loopback address never
  appears in a packet traveling across the network.

68
Summary of Special IP Addresses

• Special addresses are reserved and should never
  be assigned to host computers.
• Each special address is restricted to certain
  uses.

69
Datagram Forwarding in IP

• Forwarding is the process of taking a packet from
  an input and sending it out on the appropriate
  output.
• Routing is the process of building the tables
  that allow the correct output for a packet to be
  determined.

70
Bridges, Switches, Routers

• All forward messages form one link to another.
• Bridges are data link-level nodes and forward
  frames from one link to another (in a LAN).
• Switches are network layer nodes, which forward
  packets in a switched network.
• Routers are internet-level nodes which forward
  datagrams from one network to another.
• Bridges and switches are often called Layer2
  Switches, meaning above the physical and below
  the internet layer.

71
Datagram Forwarding in IP

• Main ideas needed to forward IP packets
• Every IP datagram contains the IP address of the
  destination host.
• The network part of the IP address uniquely
  identifies a single physical network on the
  larger Internet.
• All hosts and routers that share the same network
  part of their address are connected to the same
  physical network and can communicate by sending
  frames over that network.
• Every physical network that is part of the
  Internet has at least one router that is also
  connected to at least one other network and can
  exchange packets with hosts or routers on either
  network.

72
Forwarding An IP Datagram

• IP packets are sent from a source to a
  destination host, possibly passing through
  several routers.
• Any node (host or router) tries to determine if
  it is connected to the same physical network as
  the destination, by comparing the network address
  part of the destination address with the network
  address part of each interface address. ( Hosts
  have one address, routers have two or more, since
  they are connected to multiple networks.)
• If there is a match, the destination is on the
  same network and the packet is delivered.

73
Forwarding An IP Datagram

• If the node is not connected to the same physical
  network as the destination, it sends the datagram
  to a router, called the next hop router.
• The router finds the correct next hop by
  consulting its forwarding or routing table.
• The table is primarily a list of (NetworkNum,
  NextHop) pairs.
• There is usually a default router if none of the
  entries match the destinations network number.

74
Forwarding Algorithm

• if (Destination NetworkNum NetworkNum
• of one of my interfaces)
• deliver packet to destination over the
  interface
• else if (Destination NetworkNum is in my
• forwarding table)
• deliver packet to NextHop router
• else
• deliver packet to default router.

75
Forwarding Example
Suppose H1 wants to send a datagram to H2- on
same network sends directly. What about H1 to H8?
(R1, R2, R3)
Routing table for R2
76
Forwarding Tables

• Simple enough to be manually configured
• Usually built by routing protocol
• Routers contain tables that list only a set of
  network numbers, not all the hosts. Sometimes
  they also contain interface information.

77
Routers and IP Addressing

•   IP address depends on network address
•   What about routers - connected to two
  networks?
•   IP address specifies an interface, or network
  attachment point, not a computer
•   Router has multiple IP addresses - one for
  each interface

78
Principle of Scalability

• An important principle of building a scalable is
  to reduce the amount of information stored in
  each node.
• Most common way to do that is a hierarchical
  aggregation. IP uses a 2 level hierarchy, with
  networks at the top level and nodes at the
  bottom.
• Information is aggregated by letting routers deal
  only with reaching the right network, and the
  information that a router needs is represented by
  a single piece of information.

79
Router Implementation

• Control processor is responsible for running the
  routing protocols.
• The switching fabric transfers packets from one
  port to another.
• Routers differ from switches
• Must handle variable length packets

80
Block Diagram of a Router
81
Address Translation (ARP)

• IP addresses are virtual because they are
  maintained by software
• Neither LAN nor WAN hardware understands the
  relationship between
• an IP address prefix and a network nor
• an IP address suffix and a particular computer
• Upper levels of protocol stack use protocol
  addresses
• Network hardware must use hardware address for
  eventual delivery

82
Address Translation (ARP)

• Protocol address must be translated into hardware
  address for delivery and there are three methods
• Used with WAN hardware- uses table look up
• Uses a mathematical function to translate the
  addresses
• Uses a distributed system in which two computers
  exchange messages

83
Protocol Addresses and Packet Delivery

• An application places the data in a packet, which
  contains the protocol address of the destination
• Software in the host or router uses the
  destination address to select the next hop for
  the packet and transfers the packet.
• Both the next hop and the destination address are
  IP addresses.
• But there is a problem with this!

84
Protocol Addresses and Packet Delivery

• Problem The physical hardware does not
  understand IP addressing and addresses in a
  frame must be physical addresses.
• Solution Protocol addresses of next hop must be
  translated into hardware addresses

85
Address Translation

• Address TranslationUpper levels use only
  protocol addresses
• "Virtual network" addressing scheme
• Hides hardware details
• Translation occurs at data link layer
• Upper layer hands down protocol address of
  destination
• Data link layer translates into hardware address
  for use by hardware layer

86
Address Resolution

• Finding or mapping or translating hardware
  address for protocol address
• Called address resolution
• Data link layer resolves protocol address to
  hardware address
• Resolution is local to a network
• Network component only resolves address for other
  components on same network

87
Address Resolution
88
Address Resolution Techniques

• Three techniques are used for address resolution
  depending on the protocol
• Table lookup
• Bindings or mappings are stored in a table memory
  with protocol address as key
• Data link layer looks up protocol address to find
  hardware address
• Closed-form computation
• Protocol address based on hardware address
• Data link layer computes the hardware address
  from protocol address by using basic Boolean and
  arithmetic operatons
• Simply choose hardware address to be hostid and
  any host can determine hardware address as
• hardware_address ip_address 0xff
• Dynamic Message Exchange
• Network messages used for "just-in-time"
  resolution
• Data link layer sends message requesting hardware
  address destination responds with its hardware
  address

89
Address Resolution

• A resolves protocol address for B for protocol
  messages from an application on A sent to an
  application on B
• A does not resolve a protocol address for F
• Through the internet layer, A delivers to F by
  routing through R1 and R2
• A resolves R1 hardware address and sends packet
  to router.
• Software on R1 resolves the address for R2
• R2 receives the packet and resolves the address
  for F
• Network layer on A passes packet containing
  destination protocol address F for delivery to R1
  which forwards it to R2 and then to F

90
Address Translation (ARP)

• IP datagrams contain IP addresses, but the host
  or router to which it is sent only understands
  network addresses.
• Need to translate the IP addresses to link-level
  addresses.
• One solution is to map an IP address to a
  physical address by encoding the hosts physical
  address into the host part of the IP address.
• More general solution is for each host to
  maintain a table of address pairs and to map an
  IP address to a physical address.
• Better approach each host dynamically learns
  the contents of the table using the network and
  ARP.

91
Address Translation (ARP)

• Goal of Address Resolution Protocol (ARP) is to
  enable each host on the network to build a table
  of mappings between IP addresses and data link
  level addresses.
• Set of mappings stored in a host is called the
  ARP cache or ARP table.
• ARP takes advantage of the fact that many
  technologies support broadcast.

92
ARP Packet Format
Used to map IP addresses into Ethernet Addresses
93
ARP Message Delivery

• ARP request message dropped into hardware frame
  and broadcast
• Uses separate protocol type in hardware frame
  (ethernet 806)
• Sender inserts IP address into message and
  broadcast
• Every other computer examines request
• Computer whose IP address is in request responds,
  others discard it
• Puts hardware address in response
• Unicasts to sender
• Original requester can then extract hardware
  address and send IP packet to destination

94
ARP Message Exchange
95
ATMARP

• ARP procedure will not work with an ATM network
  because it depends on the fact that ARP packets
  can be broadcast to all hosts on a single
  network.
• On solution is to use LAN emulation, which makes
  an ATM network behave like a shared media LAN.
• LAN can be inefficient in a wide area ATM
  network.
• There is a different ARP procedure called ATMARP
  that does not depend on LAN emulation or
  broadcast.
• ATMARP relies on a server to resolve addresses.

96
Logical IP Subnets

• A large ATM can be subdivided into several
  smaller subnets, which behaves like a single
  network.
• All nodes on the same subnet have the same IP
  network number and can communicate directly.
• An advantage of the LIS model is that we can
  connect a large number of hosts and routers to a
  big ATM network with out necessarily giving them
  all addresses from the same IP network.
• This makes it easier to manage address assignment
  and improves scalability

97
Logical IP Subnets
An example of an ATP network divided into two
LIS. One has an IP address of 10 and the other
is 12.
98
Summary of Basic IP Mechanisms

• Heterogeneity-IP defines best effort service
  based on unreliable datagrams
• Uses a common packet format with fragmentation
  and reassembly
• Uses a common global address space and ARP for
  identifying all hosts
• Scalability-IP hierarchical aggregation reduces
  the amount of information needed to forward
  packets. IP addresses are partitioned into
  network and host components. Packets are routed
  first to a network and then delivered to the
  correct host on that network.

99
Host Configuration (DHCP)

• Ethernet addresses are configured into the NIC
  card by the manufacturer and assures that these
  addresses are unique.
• IP addresses, by contrast, must be unique on an
  internetwork, and also must reflect the structure
  of that network with a network part and a host
  part.
• A host also needs the address of a default
  router- the place to which it can send packets.
• Dynamic Host Configuration Protocol (DHCP)

100
Host Configuration (DHCP)

• Most operating systems provide a way to manually
  configure the IP information needed by a host,
  but there are disadvantages to this
• This is a lot of work
• It is error prone, since every host must get a
  unique number
• Usually automated methods are required, using a
  protocol called Dynamic Host Configuration
  Protocol (DHCP).
• There is at least one DHCP server that is the
  central repository for the host configuration
  information.

101
Host Configuration (DHCP)

• DHCP relies on a server that is responsible for
  providing configuration information to hosts.
• Configuration information for each host is stored
  in the server and automatically retrieved when it
  is booted or connected to the network.
• Administrator can assign addresses or allow the
  DHCP server to maintain an available pool of
  addresses that it provides to hosts on demand.

102
Host Configuration (DHCP)

• First problem faced by DHCP server is that of
  server discovery.
• To contact an DHCP server, a newly booted or
  attached host sends a DHCPDISCOVER message to a
  special IP address (25.255.255.255) that is an IP
  broadcast address.
• It is received by all hosts an routers on the
  network. (Routers do not forward these packets
  beyond this network.)
• The server would reply to the host and the other
  nodes would ignore it.

103
Relay Agent

• Since requiring a DHCP server on every network
  would need a large number of servers, the DHCP
  uses the concept of a relay agent.
• There is at least one relay agent on each network
  and it is configured with just one piece of
  information- the IP address of the DHCP server.
• When it receives a DHCPDISCOVER message, it
  unicasts to the DHCP server and waits for the
  response which it sends back to the requesting
  client.

104
DHCP
A DHCP relay agent receives a broadcast
DHCPDISCOVER message from a host and sends a
unicast DHCPDISCOVER message to the DHCP server.
105
DHCP Packet

• A DHCP packet is actually sent using a protocol
  called UDP (User Datagram Protocol) that runs
  over IP.
• The UDP packet provides a demultiplexing key that
  says This is a DHCP packet.
• Client puts its address in the chaddr field.
• DHCP server responds by filling in the yiaddr
  field (your IP address). These addresses are
  leased and the host needs to renew the lease if
  it is still connected.
• Other information such as the default router can
  be included in the options field.

106
DHCP Packet Format
107
DHCP Management

• By allowing network managers to configure a range
  of IP addresses per network rather than one IP
  address per host, DHCP improves the manageability
  of the network.
• DHCP may also introduce some more complexity to
  the network since it makes binding between
  physical hosts and IP addresses more dynamic.
• This makes the managers job more difficult when
  it is necessary to locate a malfunctioning host.

108
Error Reporting (ICMP)

• How does the Internet treat errors?
• IP drops datagrams when a fragment fails to
  arrive at a destination.
• It has a companion protocol, called the Internet
  Control Message Protocol (ICMP), that defines a
  collection of error messages that are sent back
  to the source when an router or host is unable
  to process a datagram successfully.
• Examples host is unreachable, TTL is 0, header
  checksum failed, etc.

109
ICMP Messages
110
Error Reporting (ICMP)

• ICMP also defines other control messages that a
  router can send back to a source host.
• ICMP-Redirect tells the source host that there is
  a better route to the destination.
• The source host adds this new route to its
  forwarding table and uses it for future datagrams
  addressed to that destination.

111
ICMP Transport

•   ICMP uses IP to transport an error message
•   Router creates the datagram and encapsulates
  the ICMP message in the datagram.

112
Using ICMP Messages to Test Reachability

•  An internet host, A, is reachable from another
  host, B, if datagrams can be delivered from A to
  B
•  The ping program tests reachability - sends
  datagram from B to A that A echoes back to B
•  Ping uses ICMP echo request and echo reply
  messages
• Internet layer includes code to reply to incoming
  ICMP echo request messages

113
Using ICMP To Trace a Route

•   List of all routers on path from A to B is
  called the route from A to B
•   traceroute uses UDP (User Datagram Protocol)
  to non-existent port and TTL field to find route
  via expanding ring search
•   Sends ICMP echo messages with increasing TTL
•   Router that decrements TTL to 0 sends ICMP
  time exceeded message, with router's address as
  source address
•   First, with TTL 1, gets to first router, which
  discards and sends time exceeded message
•   Next, with TTL 1, gets through first router to
  second router
•   Continue until message from destination
  received
•   traceroute must accommodate varying network
  delays
• Must also accommodate dynamically changing routes

114
Using ICMP For Path MTU Discovery

•  Fragmentation should be avoided
•  How can source configure outgoing datagrams to
  avoid fragmentation?
•  Source determines path MTU - smallest network
  MTU on path from source to destination
•  Source probes path using IP datagrams with
  don't fragment flag
•  Router responds with ICMP fragmentation
  required message
• Source sends smaller probes until destination
  reached

115
Virtual Networks and Tunnels

• On most internets, it is possible for nodes to
  communicate with other nodes on different
  networks.
• There are situations, where controlled
  connectivity s required- these are virtual
  private networks (VPN).
• Communication is restricted to take place only
  among these sites (often of a corporation),
  providing security.

116
Virtual Private Networks

• To make a private network virtual, the leased
  transmission lines, that are not shared, are
  replaced by some sort of shared network.
• A Virtual Circuit is a reasonable replacement
  because it provides a logical point-to-point
  connection between two sites.

117
Virtual Private Networks
a) Two separate private networks
b) Two virtual private networks sharing common
switches
118
Virtual Private Networks and Tunnels

• Two separate corporations may migrate to a
  virtual circuit network.
• The limited connectivity of a private network is
  maintained, but since the networks share
  switches, we say that two virtual private
  networks have been created.
• An ATM or Frame Relay can provide the
  connectivity or an IP network can be used by
  providing a tunnel.

119
Tunnels

• An IP tunnel is a virtual point-to-point link
  between a pair of nodes that are separated by an
  arbitrary number of networks.
• This virtual link is created within the router at
  the entrance by providing it with the IP address
  of the router at the far end of the tunnel.

120
Routing Through a Tunnel

• When a router at the entrance wants to send a
  packet over this virtual link, it encapsulates
  the a packet inside an IP datagram.
• The destination address is the address of the
  router at the end of the tunnel, and the source
  address is the router at the entrance.
• The virtual link, looks similar to a normal link
  in the routing table.

121
A Tunnel Through an Internetwork
1
0
virtual
R1 has two physical interfaces Interface 0
connects to Network 1, interface 1 connects to
the Internetwork and is the default. It also has
a virtual interface to the tunnel.
122
Tunneling

• Suppose a tunnel has been configured from R1 to
  R2 and assigned a virtual interface number of 0.
  The forwarding table might look like this

123
Tunneling Example

• Suppose R1 receives a packet from Network 1 that
  is addressed to network 2.
• To send it out on the virtual interface, the
  router adds an IP header addressed to R2 and then
  proceeds to forward the packet as I it had been
  received.
• R2s address is 10.0.0.1 since the network number
  of this address is 10 not 1 or 2
• When R2 receives the packet it removes the IP
  header and processes it.

124
Why Tunnels?

• Why create a tunnel?
• Greater security- it becomes a private link
  across a public network.
• R1 and R2 have properties like multicast routing
  and by connecting them with a tunnel, all these
  routers appear to be connected. This is how the
  MBone (multicast backbone ) is built.
• Tunnels can carry packets from protocols other
  than IP across an IP network. As long as the
  routers can handle other protocols, the IP tunnel
  looks to them like a point-to-point link over
  which they can send non-IP packets.
• Tunnels also provide a mechanism by which we can
  force a packet to be delivered to a particular
  place.

125
Disadvantages of Tunnels

• It increases the length of packets causing a
  waste of bandwidth for short packets.
• Routers at the endpoints must also do more work
  as they add and remove tunnel headers.
• There is also a management cost to set up the
  tunnels and and make sure they are correctly
  handled by the routing protocols.

126
Routing

• A switch or router needs to be able to look at a
  packets destination address and then to
  determine which of the output ports is the best
  one for that destination.
• In datagram networks, including IP networks,
  routing is an issue for every packet.
• In virtual circuits routing is an issue only for
  the connection request packets all subsequent
  packets follow the same path.
• The switch makes a decision by consulting a
  forwarding table.

127
Forwarding and Routing

• The fundamental problem of routing is How do
  switches and routers acquire the information in
  their forwarding tables?
• Forwarding consists of taking a packet,
  consulting a table and sending the packet in the
  direction determined by the table. This is a
  relatively simple and well-defined process
  performed locally at a node.
• Routing is the process by which the forwarding
  tables are built. This depends on complex
  distributed algorithms that continue to evolve.

128
Forwarding and Routing Tables

• Forwarding table and routing table are sometimes
  used interchangeably but there is a distinction.
• The forwarding table is used when a packet is
  being forwarded and must contain enough
  information to accomplish that task. This
  requires that a row in the table must contain the
  mapping from a network number to an outgoing
  interface and some MAC information, such as the
  Ethernet address of the next hop.
• The routing table, built up by the routing
  algorithm as a precursor to the forwarding table,
  contains mappings from network numbers to next
  hop and information about how this was learned.

129
Forwarding and Routing Tables

• There are reasons for implementing these tables
  as separate data structures
• The forwarding table needs to be structured to
  optimize the process of looking up a network
  number when forwarding a packet.
• The routing table needs to be optimized for
  calculating changes in topology.
• The forwarding table is sometimes implemented in
  specialized hardware, but this is rarely done
  with the routing table.

130
Routing and Forwarding Tables
b) A Forwarding Table- the MAC Address is
provided by the Address Resolution Protocol (ARP)
a) A Routing Table
131
Scalability

• Key question in building a mechanism for the
  Internet is Does this solution scale?
• The answer for the previous algorithms and
  protocols is NO, since they are designed for
  networks of modest size(lt 100) nodes.
• These do serve as building blocks for a
  hierarchical infrastructure that is used in the
  Internet today.

132
Domains

• These protocols collectively are called
  intradomain routing protocols or interior gateway
  protocols(IGPs).
• A routing domain is an internet in which all the
  routers are under the same administrative control
  (e.g. Single campus or single ISP)
• For now, we are considering the problem of
  routing in a small to midsize network, not the
  full Internet.

133
Network as a Graph

• Routing in essence is a problem of graph theory.
• The nodes may be hosts, switches, routers or
  networks.
• The edges of the graph correspond to the network
  links. Each edge has an associated cost, which
  indicates the desirability of sending traffic
  over that link.

134
Network Represented as a Graph
135
The Routing Problem

• The basic problem of routing is finding the
  lowest cost path between any two nodes, where the
  cost of a path equals the sum of the cost of all
  the edges on the path.
• For a simple path calculate all the shortest
  paths and store them on each node.
• Such a static approach has shortcomings
• It does not deal with node or link failures
• It does not consider the addition of new nodes or
  links
• It implies that edge costs do not change

136
Routing Protocols

• Routing is achieved by running protocols among
  the nodes. These protocols provide a distributed,
  dynamic way to solve the problem of finding the
  lowest cost path in the presence of link and node
  failures and changing edge costs.
• It is difficult to make centralized solutions
  scalable so the widely used protocols are
  distributed and are areas of challenges and
  research.

137
Distributed Protocols

• Distributed algorithms raise the possibility that
  two routers will at one instant have different
  ideas about the shortest path to some
  destination.
• Packets can become stuck in a loop if each router
  thinks the other one is closer to the
  destination. This discrepancy must be resolved as
  soon as possible.
• Assume the edge costs in a network are known.
• The two main classes of routing protocols are
  distance vector and link state.

138
Distance Vector (RIP)

• RIP ( Routing Information Protocol) dynamically
  builds a routing table using the distance vector
  algorithm.
• The idea behind the distance vector algorithm is
  that each node constructs a one dimensional array
  (vector) containing the distances (costs) to all
  other nodes and distributes that vector to its
  immediate neighbors.
• Each node knows the cost of its directly
  connected neighbors.
• A link that is down is assigned an infinite cost.

139
Distance Vector Routing

• In the next graph, the cost of each link is set
  to 1, so that the least cost path is simply the
  one with the fewest hops.
• We represent each nodes knowledge as a table.
• Note that each node only knows the information
  in on row of the table (the one in the left
  column that bears its name)
• The global view is not available at any single
  point in the network.

140
Distance Vector Routing
141
Global View of Initial Distances
142
Initial Routing Table at Node A
143
Routing At Node A

• Initially the routing table at each node reflects
  the beliefs that a packet can reach a connected
  node in one hop and that others are unreachable.
• The next step in distance-vector routing is that
  every node sends a message to its directly
  connected neighbors containing its list of
  distances.
• The router learns the new paths and can update
  its table with the new costs for next hops.
• It takes only a few exchanges before each node
  has a complete routing table.

144
Final Routing Table At Node A
145
Final Routing Tables

• The process of getting constant routing
  information to all the nodes is called
  convergence.
• There is no one node in the network that has all
  the information in this next table.
• Each node knows only the content of its own
  routing table.
• This distributed algorithm enables all nodes to
  achieve a consistent view of the network without
  a centralized authority.

146
Final Distances Stored at Each Node (Global View)
147
Other Distance Vector Issues

• When does a given node send a routing update to
  its neighbors?
• Periodic update sends every so often (several
  seconds to several minutes) even if nothing
  changes. Lets others know it is still running.
• Triggered update- sent when a node receives an
  update from a neighbor that causes a change in
  its routing table.

148
Other Distance Vector Issues

• What happens when a link or node fails?
• The nodes, that notice the failure, send a new
  list of distances to their neighbors and tables
  are updated.
• How does a node detect a failure?
• Nodes test links by sending control packets and
  wait for an acknowledgement.
• Nodes determine a link is down when it does not
  receive a periodic update.

149
Count to Infinity Problem

• Sometimes the network does not stabilize.
• ( See example p. 278) Each node advertises an
  unreachable link and the hop count increases on
  each router table in a cycle.
• Partial solution uses a relatively small number s
  an approximation to infinity.
• Split horizon solution- when a node sends an
  update, it does not include those it learned from
  a neighbor
• back to the neighbor.
• These solutions do not work for large routing
  tables.

150
Routing Information Protocol (RIP)

• Use is widespread since it was distributed with
  Berkely Unix.
• It s also simple and based on the distance-vector
  algorithm.
• Routing in internetworks differ slightly.
• In an internetwork, the goal of the routers is to
  learn how to forward packets to other networks.
• Instead of advertising the cost of reaching other
  routers, they advertise the cost of reaching
  other networks.

151
Example Network Running RIP
Router C advertises to router A that it can reach
networks 2,3 at a cost of 0 networks 5,6 at a
cost of 1, and network 4 at a cost of 2.
152
RIP Packet Format
153
RIP

• RIP is a straightforward implementation of
  distance-vector routing and one of the most
  widely used.
• Built on distance-vector algorithm.
• Routers running RIP send their advertisements
  every 30 seconds.
• A router sends an update message when its table
  changes.
• RIP supports multiple address families, not just
  IP
• It tries to find the minimum hop route.
• Valid distances are 1-15, with 16 representing
  infinity, which limits it t running on fairly
  small networks.

154
Link State (OSPF)

• Open Shortest Path First Protocol (OSPF) is the
  most widely used link-state routing protocol.
• Link-state routing is the second major class of
  intradomain routing protocols.
• Assumptions are similar to distance-vector
  routing. Each node knows the state and the cost
  of the link to its neighbors.
• Need to provide each node with enough information
  to find the least cost path to any destination.

155
Link State (OSPF)

• Basic idea Every node knows how to reach its
  neighbor and if this knowledge is disseminated to
  every node, then every node will have enough
  knowledge of the network to build a complete map
  of the network.
• This is a sufficient condition for finding the
  shortest path to any point in the network.

156
Link State (OSPF)

• Link-state routing protocols rely on two
  mechanism
• Reliable dissemination of link-state information
• The calculation of routes from the sum of all the
  accumulated link-state knowledge.

157
Reliable Flooding

• Reliable flooding is the process of making sure
  that all the nodes participating in the routing
  protocol get a copy of the link-state information
  form all other nodes.
• Basic idea is for a node to send information out
  on all of its directly connected links, with each
  receiving node forwarding it out on all its
  links.

158
Reliable Flooding

• Each node creates an update packet, called a link
  state packet (LSP) that contains the following
  information
• The ID of the node that created the LSP
• A list of directly connected neighbors of that
  node, with the cost of each one
• A sequence number
• A time to live for this packet

159
Reliable Flooding

• First two ( node ID and list of neighbors) are
  needed to enable route calculation
• Last two (sequence number and time to live (TTL)
  for this packet) are needed to make the process
  of flooding the packet to all nodes reliable.
• Reliability includes making sure that you have
  the most recent copy of the information, since
  there may be multiple contradictory LSPs.
• Making the flooding reliable is quite difficult.

160
Link State Packet Flooding
a) LSP arrives at node X
b) X floods LSP to A and C
c) A and C flood LSP to B but not X
d) Flooding is complete
161
Link State Packets

• Like RIP, each node generates LSPs
• When a periodic timer expires
• When there is a change in topology
• The newest information must be flooded to all
  nodes as quickly as possible, while old
  information must be removed and not allowed to
  circulate.

162
Goals For LSPs

• Minimize the total amount of routing traffic
• Avoid generating LSPs unless necessary by using
  very long timers. Assume messages saying no
  change do not need to be sent often.
• Make sure that old information is replaced by
  newer information by inserting sequen