CS791: A Critical Look at Network Protocols

Lecture date: 09/28/99

Prof. John Byers

Scribe: Khaled Harfoush

 

 

Routing Vs. Forwarding:

-Routing Algorithms:

    1. How to share routing information:
      1. Distance-Vector:

        2. Vectors can be huge. Vector size = O(N)

        N : the number of networks.

      2. Link-State (Shortest Path Forwarding):

Link-State Packet size = O(1) (Because packet contains only

information about its direct neighbors within its network)

Number of Link-State Packets= O(N) (Because packets must be

sent throughout the network – reliable flooding).
    1. Building Tables: (Dijkstra)

-Forwarding:

Choosing right output port given routing information.

 

 

Intra-domain routing: within an AS

  1. RIP:

    2. Is a distance vector algorithm. Works well for small domains.

  2. OSPF:

Is a link-State algorithm. Better for larger domains.

 

Inter-domain routing:

Is harder. Why?

1.

Each router in an AS has to keep all information? Scalability?

Key router in an AS keeping track of all information?

Failure of centralized node?

2.

Different metrics on different ASs.

3.

Peering agreements between ASs? Security? How much flow can go from

an AS to another? Economics and bandwidth considerations.

4.

Obfuscation. Network providers do not want to advertise their AS topology.

5.

Decentralized administration? Who is responsible for managing all this?

Fixing problems? (e.g. routers misbehaving)

  1. Exterior Gateway Protocol (EGP):

    2. Ran by border routers; so they can communicate with each other even the ones

    on the border of same AS. ( Using TCP reliable communication).

    Network reachability info.

    Peering relationships.

    Initially: routers know the networks inside their AS and advertise ASs that

    they are directly connected to. Works like link-state.

     

  2. Border Gateway Protocol (BGP):

    3. One speaker for an AS – maintains routing tables and transmitted information.

    Problem is that the BGP table is huge.

     

  3. Wide-Area routing protocols:

(Same administration is a requirement)

Landmark hierarchy:

Node has radius i. All nodes within that radius View this landmark.

Routing tables are smaller and the number of forwarded measures is near optimal.

The radius distribution is heavy tailed (A small number of routers have a large radius).