C SC 481.20 Lecture 11 : Routing Algorithms and IP Routing
major resource: Computer Networking (4th Edition), Kurose and Ross, Addison Wesley, 2008

[ previous | schedule | next ]

Routing Algorithms

Overview

• Use graph theory to present algorithms
  • nodes : routers (are assigned IDs)
  • edges : links between routers (can be directed or undirected)
  • weights on edges: any cost measure of using that link (distance, congestion, capacity)
  • shortest path : minimum number of links (hops)
  • least cost path : minimum total cost, regardless of # hops
• Static vs. Dynamic routing algorithms
  • static means routes change very slowly (e.g. manual changes to tables)
  • dynamic means routes evolve as network loads and topologies change
• Global vs. Local routing algorithms
  • global means all nodes have complete knowledge of topology and costs (e.g. link state)
  • local means nodes receive knowledge only from neighbors (e.g. distance vector)
  • both are distributed in the sense that each node does its own calculations
  • truly centralized algorithm is impractical -- why?

Link State Algorithms, illustrated by Dijkstra's

• Before algorithm is run, all nodes must have complete cost and topology information.
• Knowledge achieved by each node broadcasting costs to get from itself to its neighbors.
• Algorithm works for either directed or undirected graphs.
• Undirected means assume link cost is same in both directions (not always true).
• Algorithm calculates shortest path from a given source to all destinations.
• Each node runs algorithm with itself as source.
• Results of this iterative algorithm go into routing table.

Distance Vector Algorithms

• Each node communicates only with its neighbors
• Each node calculates only its own table
• Distance vector table layout:
  • Each row is a Distance Vector received from neighbor Y
  • Each column is a destination router X
• Distance Vector from neighbor Y contains cost from Y to every other node (forms a vector)
• Each entry represents cost to destination X via neighbor Y.
• Least cost path for each destination is copied into forwarding table.( dest., neighbor, maybe cost)
• Table is dynamically updated. When? When node receives message from neighbor node.
• When does node send message? When its own update causes change(s) to least cost path.
• What does message contain? Destination ID and cost (note: neighbor ID is not in message).
• “good news travels fast” – when a link cost decreases, full reduction propagates rapidly (one hop per update)
• “bad news travels slow” – when a link cost increases (or link goes down) the increase may propagate at only one cost unit per update (the “count to infinity” problem). There are countermeasures.

Hierarchical routing

• How large would routing tables be if every router needed an entry for every other router and host?
• Define autonomous systems (AS) as logical grouping of routers.
• Inter-AS routing is routing from one AS to another.
• Intra-AS routing is routing within an AS.
• They can use different routing methods.
• ANALOGY: Post office as AS. use 5-digit ZIP code to deliver to destination post office (“inter-AS”), street address to deliver to host from there (“intra-AS”)
• Routers at the AS “edge” are called gateway.
• Gateways communicate with other gateways to implement inter-AS.
• Gateways must be able to run routing methods for both inter- and intra-AS.

Routing IP datagrams

RIP – an intra-AS protocol

• Routing Information Protocol
• A distance vector protocol
• Distributed with early TCP/IP as part of BSD unix
• Cost is hop count (e.g. each link has cost 1).
• 15 hop limit.
• Distance vectors exchanged every 30 seconds
• Exchange protocol uses UDP and port 520. Yes, transport layer protocol used to implement network layer function!

OSPF – an intra-AS protocol

• Open Shortest Path First
• A link state protocol
• Uses Dijkstra’s algorithm
• Has security features (exchanges authenticated)
• Allows use of multiple same-cost paths
• Can use different cost metric for different TOS, e.g. different graph for each TOS
• Supports hierarchical structure w/i the AS.
• AS partitioned into areas, each area has border router, one area has backbone.
• Backbone routes among area border routers
• Area border routers route within area

BGP – an inter-AS protocol

• Border Gateway Protocol, RFC 1772
• De facto standard for Internet
• For routing between AS’s. E.g. between organizations.
• Policy and political considerations now come into play (e.g. routers on AT&T networks will not route through routers on MCI networks).
• Path may not be “optimal cost”
• “Path vector” algorithm, a variation of distance vector
• Routers maintain path to each destination AS. In BGP, these paths are periodically communicated to neighbor (called “peer”).
• Note that exchanged paths do not contain cost information.
• Example:
  • network with routers A through H.
  • A is a peer of B
  • B has this route for datagrams destined for H: BCEH.
  • B sends this path info to A.
  • A then may set its path to H as: ABCEH, but doesn’t have to.
  • Decision not to, may be technical (already has shorter path) or political (A will not route anything through E because E is owned by competitor).
• Routers exchange BGP messages via TCP, port 179.
• There are 4 types of BGP messages:
  • open – to establish contact with peer
  • update – to send (“advertise”) path vector to peer
  • keepalive – let peer know your connection is still OK
  • notification – let peer know that error has occurred
• BGP can also be used for intra-AS routing (IBGP, Internal BGP)