North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: Verio Peering Question

  • From: R.P. Aditya
  • Date: Tue Oct 02 22:34:40 2001

On Tue, Oct 02, 2001 at 09:58:56PM -0400, Jeff Mcadams wrote:
> Is your position, honestly, that a provider the size of IgLou doesn't
> *deserve* to have robust network connectivity because we're not big
> enough?  If so, I'll be sure to avoid purchasing any transit from you.

Aggregation vs. leaking specifics is about hierarchy -- the barrier of entry
to announcing routes into the global BGP tables is very low -- a relatively
small provider, with 1 router, in some corner of the world can cause bloat to
the *global* tables and thereby force every other provider to share the cost
of accomodating that bloat (likewise "big" providers who don't aggregate are
also culpable, but also causing themselves pain, but they get compensated
directly for that). Prefix length filtering is an interimn fix, an attempt,
like import-duty to provide protection.

> Also sprach Sean M. Doran
> >| Oh, and that scales really well.</sarcasm>
> 
> >The solution to this is very simple: aggregate.

The entire ?LA scheme proposed for IPv6 smacks of the hierarchy; that those
who build coast-to-coast or worldwide networks have _more_ of a right to
introduce prefixes into the global table because they can aggregate better,
and to some extent, that is a reality, that's the way it's going to have to be
if the DFZ is going to be of a manageable size with current protocols.

Unfortunately, even with a clear hierarchy, aggregation hides a lot of
information about a given AS to peers and hot-potato routing forces even the
bigger (in terms of total prefix space) providers to leak specifics to manage
their traffic, but that can be worked out on a per-peer basis.

*Luckily*, for the little guys, the "right" to multihome throws a wrench into
the original IPv6 ?LA hierarchy model. However, so far, there is no realistic
model that allows for a limited DFZ AND allows the leaking of specifics as
required per the IPv4 multihoming model.

I'm betting that a multiple address (prefix) host model, with something like
SCTP will likely be the longterm "solution". Painful, but it moves the tragedy
of the commons away from the network layer...

Adi