I'm inclined to think that some smart folks might ought
to think about whether we want to keep going with Path-Vector,
or whether some other algorithm (whether DUAL or something
completely outside the Bellman-Ford class of algorithms) might
have better properties. I am decidedly not a mathematician,
so I'm only urging that clever folks mull this over. It
might also be interesting, from an academic perspective at
least, to consider whether traditional EE control theory has
anything to offer us regarding inter-domain routing.
I'm also inclined to think that some of the overloading
in BGP isn't helping the overall stability of the routing system.
If I were more awake (or had had coffee recently), I could likely
create a long list of these. Instead, I'll select one *example*:
- right now a number of ISPs prepend their AS several times
("AS padding") as a method of the originating AS to
make their path to a customer less preferred than
some other ISP's announcement of the identical customer
prefix.
- we could consider adding a variable to denote that a
given path is less preferred so that the above practice
weren't needed.
Finally, and perhaps most radically, it might be helpful
or clever to consider partial separation of the "topology
discovery" mechanism from the "policy communication" mechanism.
Right now, both are intermingled in BGP. Both are useful. The
question is whether it might be useful to have both exist, but
exist separately and distinctly.
I'm afraid I've more questions than answers,
so I'll end here for now.
Cheers,
Ran
rja@inet.org
This archive was generated by hypermail 2b29 : Mon Aug 04 2003 - 04:10:04 EDT