I've heard this from various networks as well. While I find it
annoying, it does serve usefull purposes. More peering sessions mean
higher CPU load. higher CPU load can mean more instability within the
router. When the router crashes or otherwise reboots, it takes it
longer to resynchronize which means longer delays in convergance and
poorer overall service. Limiting the number of peering sessions on
really busy routers is a benefit, not a problem. The only thing that
could be asked of them would be that they investigate installing a
second router to offload some of the peering sessions so they don't
have such a backlog of requests and maybe provide load balancing. if
they are accepting applications, they need to provide facilities for
those requests to be implemented.
However, I still second the comment below..
Has anyone actually studied how much load adding a peering session is? How
much is added if the peering session is via the RS (I.e. how much does it
add to hear X routes from a new source). This seems like something that
would be nice to know.
Sure this would be a great idea if you could get them to buy off on the idea
that peering is worth a second or third router (~80k router + ~5k
port/month). My question is "What about the RA?". The whole idea was to
remove the load issues, flap and dampening problems, etc from the routers and
move it to a CPU only box. Granted there are other things to look at, which
we don't need to get into with this thread but at least it would be a
temp solution. IMHO.
BTW Whoever sent that mail about BBN peering. They just added two more
routers to relax the CPU issues at MAE_EAST. You may want to email
Todd R. Stroup
Fiber Network Solutions, Inc.