From: Scott Ramsby (sramsby_at_cs.washington.edu)
Date: Tue Feb 03 2004 - 12:49:44 PST
I thought this over again and I'm not sure we can eliminate the
discovery pings. Those can be used to gather information slowly about
our neighbors via their linkstate broadcasts and such, but not to drop
neighbors. We could time-out neighbors I suppose, but we have no idea
how long it might be before they have more information to advertise or a
packet to forward. Do we just assume that there will always be periodic
traffic and set our timeout to REALLY long or what? I realize in a real
network there will always be lots of traffic, but in our P2P thing,
without manual pings I have to rely on the other node's LinkState
broadcast timers, which I have no knowledge of the frequency.
Scott
Janet Davis wrote:
>Hi all,
>
>Abhinav and I just discovered an important interoperability issue with
>neighbor discovery and link state routing.
>
>The scenario is this:
>
>* For neighbor discovery, node 1 sends a ping with a special message and
>TTL=1, and expects to receive a ping with the same message back.
>
>* Node 2 receives this ping, discovers it has no route to node 1, and
>consequently does not send a reply.
>
>* As a result, node 1 never recognizes node 2 as its neighbor, and since
>we only use bidirectional links in our Dijkstra computation, node 2 never
>does compute a route to node 1.
>
>For interoperability, we all need to agree on one method of recognizing
>your neighbors. The most efficient approach is to use the 'from' argument
>of the onReceive method -- it's guaranteed that this is the address of one
>of your neighbors, regardless of the packet contents.
>
>Note that if you use this method, you can eliminate the special neighbor
>discovery ping packets entirely! (Why?)
>
>So, for this assignment you should decide who your neighbors are based on
>the 'from' argument of the onReceive method, not on the contents of
>neighbor discovery ping packets.
>
>Cheers,
>Janet
>
>
>
_______________________________________________
Cse461 mailing list
Cse461_at_cs.washington.edu
http://mailman.cs.washington.edu/mailman/listinfo/cse461
This archive was generated by hypermail 2.1.6 : Tue Feb 03 2004 - 12:48:21 PST