Re: routing protocols will provide path-MTU

[Steve Deering <[e-mail]>]

Craig and Jeff both mentioned one of the possible ways in which routing
protocols might provide path-MTU information to hosts, by disseminating
MTU info as part of their routing updates.  (They mentioned link-state
algorithms.  I believe cisco's IGRP also supports this.)  However, as
they observed, this doesn't work with information-hiding, subnetting,
default routes, etc.  Basically, it doesn't scale -- it is unreasonable
to expect your neighboring gateway to know the path-MTU to every possible
Internet destination (and for every TOS).

Another possible approach would be one in which a host asks a first-hop
gateway for the path-MTU to a given destination, and the gateway then
goes out and probes the path, all the way to the last-hop router, and
returns an answer to the host.  I don't see that this offers any
advantages over the schemes we have been discussing; it has the
drawback of forcing a full RTT (from source host to last-hop-router)
delay before sending data, unlike the RF-bit scheme (no delay) or the
DF-bit scheme (no delay if no MTU shrink point, one RTT delay to the
shrink point if there is only one, multiple increasing RTTs if multiple
shrink points).

A third possibility is that the "coming revolution in routing protocols"
will impose the need for all connections/session/associations/whatever
to start with a route setup phase (e.g., passing authenticators,
setting up policy paths, etc.), and the MTU will be provided as part of
the route setup.

What I was trying to find out was, when it was said at the IETF meeting
that, "Within 'a few' years, the routing protocols will provide path-MTU
information, so MTU discovery will be unnecessary.", what did the
speaker(s) have in mind?  And why should we believe them?

Steve