Re: Intra-AS Routing Standards

[hedrick at cs.rutgers.edu]

I'm going to comment on this here, because I'm not going to be at the
next IETF, and so won't be able to partipate in discussions there.  I
agree with the goal of having a single recommended protocol, and with
much of the proposed procedure, but I do have one substantial problem.
The emphasis is on making decisions via discussions at IETF.  I claim
that the decision should actually be based primarily on an evaluation
of field experience and (if available) simulation work.  Routing
protocols are rather different than the other protocols IETF deals
with.  In telnet, it's pretty clear what the effect of a new option is
going to be.  With routing, it is not at all clear what the overall
system impact of a particular design is going to be.  I am
particularly concerned with whether protocols are going to stand up to
problems such as flapping lines in a large network.  Thus I would like
to see this IETF concentrate, not on making a decision, but on
starting a datagathering process and setting up criteria to be used
for making the decision.  I suggest that what you need is the
cooperation of network managers in several large networks.  They
should try to collect data over the next few months on things such as
how much of their CPU is going to the routing process, whether they
are seeing routing loops, what happens when there is a flapping route,
etc.  If you make the decision based on the vote in a mass meeting,
you are guaranteeing a decision made on emotional, not technical,
grounds. 

Note by the way that although I'm chairman of the ODV group, I'm
relatively unbiased in this discussion.  Our group does not have a
protocol to offer for this process.  IGRP was rejected by the group at
the last IETF because of patent issues.  (I'm not sure I approve of
that decision, but there was a very clear message.) As far as I can
tell, the group members do not have the time available to produce a
new protocol anywhere near soon enough for consideration. 

Finally, I'd like to see more effort go into evaluation of routing
technology, both by looking at how routing is going in the current
internet (probably in an IETF group) and by encouraging projects in
analysis and simluation of the behavior of routing protocols in
complex networks.  Let me say again: the biggest problem in routing is
not lack of protocols.  It's lack of understanding about how those
protocol behave.  When discussing distance vector vs.  SPF algorithms,
it because very clear that all we have are people's "gut feelings"
about how things are going to work out.  Jose Garcia-Luna did a nice
simulation, but unfortunately it simulated somewhat simplified
versions of the algorithms, without many of the stability features and
other complexities that actual implementations have.  I'd like to see
this work extended to include more realistic routing algorithms.
Indeed my preference would be to require that people submitting
proposed routing protocols must include realistic simulators, so that
those who wanted to evaluate the protocol could try them out on a few
networks.  I don't propose that we should hold the current decision
for this, but we should start thinking about it for future routing
protocols.  However I feel strongly that even the current decision
should at least have significant field experience, with a structured
mechanism for gathering information about that field experience.