[e2e] Overly Overlay; Peer to peer is commonplace

J. Noel Chiappa jnc at ginger.lcs.mit.edu
Wed Jan 2 07:38:54 PST 2002


    > From: "David P. Reed" <dpreed at reed.com>

    > I meant what I said, which does not contradict what you say in the
    > least. Though you apparently read it that way, I was not arguing for
    > some kind of naive "routing based on GUID" scheme.

Well, you didn't give any detail, so I was kind of at sea as to how to read
it. So, I didn't assume that's what you meant. In fact, even with this
follow-up, I'm still kind of confused as to exactly what it is you are saying.

(In my reply, I just tried to state what I believed to be the limits - and
indicate that I was OK with non-topological names for some purposes.)


    > To amplify my own thoughts in this area: topology is what routing is
    > about, and routing calculations need names that reflect topological
    > information.

OK so far.

    > However it is a deep mistake to assume that the proper model for the
    > ultimate Internet is dominated by a fixed and unchanging graph of links
    > and nodes.

Nobody's saying it's fixed and unchangeable: if nothing else, new stuff gets
put into service, and old stuff removed, all the time - and that's something
we already deal with.

However, there is a question of the dynamicity/rate. A topology which is too
dynamic (in the provisioning sense) is no more tractable than a topology that
is too dynamic (in the up/down sense). The latter is a circumstance we've
already had a lot of struggling with. Which is not to say that there's no way
to handle it - we manage to deal with a very large amount of up/down
dynamicity by constraining the scope of the updates.

However, if the topology is provisioning-dynamic, since the naming
abstraction hierarchy has to be based on the the topology, this could cause
problems.


    > Wireless networks, especially densely scaled mobile wireless networks,
    > do not behave like "wires without wires" or "fibers without fibers".
    > Topology is not naturally hierarchical in its interconnection, for
    > example.

Nobody is saying it is. But you have to impose a naming abstraction
hierarchy, or the routing will not scale. You can impose such an abstraction
hierarchy on *any* graph - it doesn't have to be hierarchically organized.
Kleinrock/Kamoun shows that when you do so - in an arbitrary graph - you
still get reasonably optimal paths (which asymptote to the optimal as the
network gets large).

    > So "hierarchically derived" topological addresses are just plain wrong.

Sorry, I completely disagree - although perhaps I'm just misunderstanding
what you mean by ""hierarchically derived topological addresses".


    > More relevant, though again as naive as GUID-based routing, is
    > geotemporal routing.

Sorry, I don't know what that is - can you elaborate?

	Noel



More information about the end2end-interest mailing list