Routing: Cores, Peers and Algorithms

1
Routing Cores, Peers and Algorithms

• Chapter 14

2
The Origin of Routing Tables

• Routers form the base of an internet and handle
  all traffic except for direct delivery from one
  host to another
• We have two questions
• What should be in routing tables?
• How do routers get the information for the
  tables?
• Initial routes may be determined
• from secondary storage into main memory at
  startup
• from a set of addresses of connecting networks

3
The Origin of Routing Tables

• In small internets, the routing tables may be
  established and modified by hand
• In large internets, automated approaches are
  necessary

4
Routing with Partial Information

• Hosts usually have two routes in their routing
  table
• A route for the local network
• A default route for nonlocal datagrams
• Example of road signs with partial information,
  page 255
• Extreme Architectural Approaches
• Star shaped topology with one road to each town,
  one central point
• Arbitrary roads with signs for all towns at each
  intersection

5
Routing with Partial Information

• The two extreme architectural approaches fail
• Star - one machine would have to work as switch
  between all
• Arbitrary - to keep all routing information at
  all sites is cumbersome and difficult to change
• A third approach
• Half of the cities lie in the east and half in
  the west
• A bridge spans a river that separates the east
  and west
• Road signs in the east list all destinations in
  the east, and only points to the west, and so on

6
Original Internet Architecture and Cores

• When TCP/IP was developed, research sites were
  connected to the ARPANET (Internet backbone)
• Routing tables were initialized and modified by
  hand
• To begin automation for routing,
• a small central set of routers kept complete
  information about all possible destinations
• a larger set of outlying routers kept partial
  information
• allows local administrators to make local changes
• default routes at the outlying routers could
  indicate a central intersection

7
Core Routers

• Those early routers were either
• Core routers controlled by Internet Network
  Operations Center
• or Noncore routers controlled by inividual groups
• Core routers communicated among themselves
• information was consistent and the system was
  reliable
• Sites assigned an Internet network address agreed
  to advertise that address to the core system

8
Core Routers

• The early Internet core routing system consisted
  of routers connecting local area networks to the
  ARPANET
• See Figure 14.1
• Hosts on the local networks passed nonlocal
  traffic to the core router
• Routing information was exchanged between core
  routers
• Default routes were not used would be inefficient

9
Core Routers

• This approach became impractical because
• the Internet outgrew a single, centrally managed
  long-haul backbone
• protocols needed to maintain consistency among
  core routers became nontrivial
• maintaining correct routing information became
  difficult

10
Peer Backbones

• NSFNET attached to ARPANET through a single
  router in Pittsburgh
• The core had explicit routes to all destinations
  in NSFNET
• Routers inside NSFNET knew local destinations and
  used a default route (the Pittsburgh router) to
  send all non-NSFNET traffic to the core
• Multiple connections were added between NSFNET
  and ARPANET as shown in Figure 14.4 and the two
  became peer backbone networks, or peers

11
Peer Backbones

• With such an architecture, how do we route with
  peer backbones?
• Routing loops are possible as in Figure 14.5
• Core systems work best for internets with a
  single, centrally managed backbone
• Expanding the topology to multiple backbones
  makes routing complex

12
Automatic Route Propagation

• The original Internet core system propagated
  complete routing information to all core routers
• Today, many routers continue to communicate
  routing information
• We need ways to automatically determine routes,
  and to continually update the information needed
  to determine the routes

13
Distance Vector Routing(aka Bellman-Ford)

• A router keeps a list of all known routes in a
  table
• When it boots, the routing table is initialized
  with an entry for each directly connected network
• Each entry identifies a network and the distance
  to it, usually in hops - See Figure 14.6
• Periodically, all routers send a copy of their
  routing table to any other router it can reach
  directly
• If a shorter path is found in one of these, the
  current path is replaced with the shorter one

14
Distance Vector Routing

• See Figure 14.7 which shows an existing table for
  router K, and an update from router J
• Is this a good choice?
• Easy to implement
• If routes change rapidly, routes may not
  stabilize
• Some routers may have incorrect information
• Message size is proportional to the number of
  networks in an internet

15
Gateway to Gateway Protocol

• GGP was used by original core routers to exchange
  routing information - no longer used
• Designed to travel in IP datagrams like TCP and
  UDP
• Messages had a fixed format header identifying
  message type

16
Distance Factoring

• Distance values were small, and the same values
  tended to be repeated often
• To reduce the message size, avoid sending copies
  of the same distance number
• The list of pairs (Network, Distance) is sorted
  by distance
• each distance is represented once and all
  networks at that distance follow

17
Reliability

• Most routing protocols use connectionless
  transport
• Routing protocols that use UDP or IP must
  compensate for failures
• by using checksums
• sequence numbers for out of order delivery

18
Link-State Routing

• Primary alternative to distance vector routing
• Tests the status of all neighbor routers
• Short message asking if neighbor is alive
• If the neighbor responds, the link is up, else
  down
• Routers periodically broadcast a message with the
  status of each of its links
• Indicates whether communication is possible
  between pairs of routers
• When link status information arrives at a router,
  it modifies its view of the internet

19
Shortest Path First

• When link status changes, the router recomputes
  routes by applying Dijkstras Shortest Path
  algorithm
• SPF computes the shortest paths to all
  destinations from a single source
• Advantages
• each router computes routes independently
• messages are propagated unchanged
• size does not depend on number of networks in
  internet

20
Summary

• Hosts and routers usually contain partial routing
  information
• The Internet used a core routing architecture in
  which cores had complete network information
• Routing information is exchanged periodically
• using distance-vector or link state algorithms

21
For Next Time

• Read Chapter 15