routing meltdown - let's talk at Pittsburgh


Could you please sign up to lead a discussion of the impending routing
meltdown as you see it at the upcoming NANOG meeting. You've been
very vocal on CIDRD, but it is not clear what solutions you are

I sincerely would like to know what you feel the problem is (though I
have a good handle on this part already, I think) and what could be
done about it.

Specifically (for discussion at the meeting):

  1. Number of destinations - 30K one path is beyond the technology
      available today, but it is pushing what parts of the Internet can
      support. Right?

  CIDRization helps here (and in 2 and 3 below). How can we
  better promote CIDRization? Can we do a better job of
  identifying who still needs to CIDRize? Are we ready to do
  proxy aggregation, and if so on what scale? Under what
  circumstances is hostile proxy aggregation (without the
  approval of the originator of the routes) justified? What
  safeguards should be in place to prevent hostile aggregation
  with no approval whatsoever? Or should the ability to do
  hostile aggregation be a feature???

  2. Number of paths - what is the limit on paths 100k, 200k? Are
      there ways to reduce the number of paths at hot spot routers?

  a. If the hot spot routers are at the exchanges and the
  problems are too many peers would it help to take routing from
  a single third party like a route server? Is there something
  wrong with the route server model, or the people running the
  route server? (save RA flames for the meeting please!)

  b. If the problem is peering at too many exchange points with
  the same people, would it help to limit your peerings with
  other major providers to three or even two exchange points?
  This would need to be configured, so would some sort of
  registry for this type of configuration help?

  3. CPU load. Are the efforts of Cisco in implementing route flap
      dampenning paying off? If so, do we anticipate that this will
      solve the problem or just buy us time (at this point it may be
      very preliminary)? Will the RA RS support dampenning?

If you prefer, I will ask ANS to rent me an asbestos suit and I'll do
my best to lead a discussion on this.

This is just a suggested outline. If you want to lead the discussion,
of course you are welcome to structure the discussion as you see fit.
I'd like to hear your assessment of what the problems are and how you
think the Internet community should be addressing them.

There are people who claim that they are not opposed to the idea of
documenting the routing topology to better coordinate routing
configuration, particularly aggregation, but object to the way it is
currently being done. I'd like to know how it could be better done.
There is also the sticky issue of who will do it, whatever this better
plan happens to be.