[c-nsp] 6504-E crash after bringing up lots of BGP sessions

[Andy B.]

Hi,

I am facing semi-random reloads of one of my routers when it is under
heavy load while receiving lots of routes from its BGP peers.

I run several 6504-E in my backbone, all with the very same IOS and
all interconnected for the same purpose: Edge Routers for BGP peering
/ customers and transit.

All routers are fully meshed (next-hop-self) and there is also a route
reflector (quagga) talking to every router.

Every BGP peer has its own route-maps for various reasons like
communities, prepends, ...

Recently I had a fiber cut to one of these routers and it had lost
connectivty to all other inernal routers. When the fiber cut was fixed
the routers started to reannounce their prefixes to each other. After
a while being at 100% CPU, the router reloaded itself without giving
any piece of information.

This happened more than once and it seems to happen when there is a
massive flood of prefixes coming in. I am not sure how to explain this
otherwise. I have other routers with much more peers and customer
links and they don't appear to have this reload issue. I am aware that
lots of route-maps will cause the CPU to remain at 100% for several
minutes and I can live with that, but I cannot live with random
reloads.

More information about the router:

#sh ver
Cisco Internetwork Operating System Software
IOS (tm) s72033_rp Software (s72033_rp-IPSERVICESK9-M), Version
12.2(18)SXF15a, RELEASE SOFTWARE (fc1)
Technical Support: http://www.cisco.com/techsupport
Copyright (c) 1986-2008 by cisco Systems, Inc.
Compiled Tue 21-Oct-08 01:14 by kellythw
Image text-base: 0x40101040, data-base: 0x42DD70D0

Changing to a different IOS (tried SXI3) did not change anything - in
fact it caused the router to give up even faster.

Could this be a memory issue? How would I be able to find that out?


Thank you for any help.

Andy