> From: Tony Li <[email protected]>

    > Well, to me, central to the notion of the Internet architecture and
    > qmore specifically to a routing architecture is the notion of topology.
    > It would seem somewhat challenging to be able to describe topology
    > without being able to describe a graph.

I agree pretty much 100% with all you've said, but one minor caveat (and the
credit for this observation goes to someone else, but I can't remember who,
or what mailing list - sorry): using a graph as a model of the network
connectivity (probably a more accurate term than 'topology', but I think
we're stuck with the latter now) usually works very well, but... it is not
necessarily the be-all and end-all. For some networks (e.g. wireless) it may
not work so well. Graphs do have issues in other cases too, e.g. this one:

  http://mercury.lcs.mit.edu/~jnc/tech/abstraction_flame.html

Having said that, I don't have what I consider a superior alternative, for
the mathematical model of network connectivity, here in my hip-pocket, ready
to offer. Like a like of things in routing, there are trade-offs: accuracy
versus amount of information needed to describe a particular connectivity
pattern is one, and for _most_ networks, using a graph to model the
connectivity does really, really well on this.

I don't know how difficult it would be to come up with a routing architecture
that could support multiple models simultaneously (e.g. a graph for most of
the network, but other stuff for local regions, e.g. perhaps something
geographic for a wirless network, etc, etc). My intuition says it would
probably be hard (impossible) to design an architecture where one could
incrementally deploy new models in an interoperable way.

But all that is pretty advanced - we're still a long way from an architecture
which would allow even simpler stuff, like multiple path-selection algorithms
with incremental deployment of new ones, local control of abstraction, etc. (I
mean a long way in deployment terms - a vaguely plausible design was done a
long time ago.)

        Noel
_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg

Reply via email to