How trustworthy are existing BGP AS Paths?

Dale,

There is a few things about it. One is that if two ASes speak an IGP
that cannot carry ASpaths (like the most used ones today) you loose
the first AS if the second announces this in an EGP somewhere. Of
course one only cares about this if the first one also announces this
using an EGP somewhere. In cases where the first AS never announces
this with an EGP, the AS is invisable and cannot be seen anywhere, so
it may as well not exist.

Second is the misconfigurationwith any network that connects two or
more ASes. 
If both announce this, it'll appear as originated from both ASes. This
is true for any link between two or more ASes (which includes DMZs).
This in my view is misconfiguration, or you could argue that this
network really is in multiple ASes, or perhaps even should not be
routed. 

I can live with the second case, but in the first case I think this is
weird network engineering. If there are two ASes they should run an
EGP between the two, be it EGP itself or whatever. In the old days
people would run IGPs between ASes because most IGPs have a better
convergence times, but that no longer holds for current EGPs.

The problems with AS paths (and the validity of them) has always been
our problem with using paths in the ripe-81 context. However, the
thing you mention is only about originating ASes and not necessarily
concerns paths. If you want to do path based policy, you have to know
a lot about the topology of the network, and you give away control of
part of your policy. Because if you base a policy on a path that
contains ASes that are some AS hops away, and that AS decides to
change its config, you may end up loosing connectivity. However, we
have had this discussion many a time, and it is mentioned more than
once in ripe-81++.

 *   Also, AS-Path expressions are presumably not useful for expressing
 * policy about EPG speakers.  (?)

Depends. Stubs EGP speakers are fine, as long as all next ASes along
the way speak an EGP that can carry paths.

 *   All in all, if these things are true, do folks here have feelings
 * about how advisable it is to implement policy in terms of AS-path
 * expressions in 4Q94?  Guesses about 3Q95?

I don't care really. People who specify their policy in terms of paths
should know damn well what they are doing. If they don't, perhaps they
deserve to be misconfigured because of it ....

-Marten




-------- Logged at  Sat Oct 1 00:58:07 MET 1994  ---------