[ih] state of the internet probes? (was Re: AOL in perspective)

Noel Chiappa jnc at mercury.lcs.mit.edu
Fri Sep 19 06:22:35 PDT 2025


    > From: Karl Auerbach

    > I think Noel C. is present on this list.

You rang? :-)

    > How does this new work dovetail, if at all with the Nimrod idea?

It's interesting that source routing is being used significantly for
traffic engineering now. (I'll try and keep this short, but it needs to
have enough detail to convey a complex set of thinking.)


I should explain (I'll get to the 'source routing' below) that there were
two main foundation stones in Nimrod: i) the use of maps as the
fundamental data to be exchanged, and ii) the use of "unitary path
selection" (where a single entity selects the entire path - perhaps using
local agents, recusively, for sub-sections of the path), not 'hop by hop
path selection', as was until recently used almost everywhere in the
Internet.

That was all driven by a incredibly long list of needs/desires: some of
them from the users, e.g, policy-based routing (but which turned out to
be something that the users didn't _really_ need, at least in the
short run); some perational, e.g. avoiding routing loops. (Traffic
engineering would be in the first category, of course.)

We didn't think that _all_ those capabilities would really be desired,
eventually; but we didn't have a perfect crystal ball, so we couldn't
ignore the 'irrelevant' ones. (Which ones are they?) The charm of the
map/unitary approach was that it was so _incredibly flexible_ - so that
whatever needs sprang up eventually, we'd be able to deal with them.
(Important if you're trying to do the architecture of a communication
system with a very long lifetime.)

(For example, if people eventually build a quantum communication network,
where qubits are sent through the network - Nimrod can handle that!)


Anyway, source routing. When you do unitary path selection, how do you
get the traffic to take that path? One obvious option is source routing
(so source routing appeared in Nimrod not as a goal itself, but as a
consequence of deeper things - the unitary path selection). But source
routing has two downsides: i) you have to carry the source route in all
the packets; ii) forwarding source routed packets is more complicated.

So Nimrod eventually explored other approaches, such as installing paths
(which eventually led to MPLS) - which also had downsides, which we
looked for ways to mitigate. All leading to even more complexity. Maybe
source routing would have been the right choice after all.


It will be interesting to see how the struggle between goals and
complexity eventually settles out (alas, I won't get to see it).

	Noel



More information about the Internet-history mailing list