[c-nsp] BGP confusion in 12.2(18)SXE2

[Roman Sokolov]

Hello Gert,

Wednesday, September 14, 2005, 11:20:49 AM, you wrote:
GD> Just spoke to some other colleagues that do not see the problem on
GD> SXE1 - so it seems to need some specific circumstances or "trigger".

GD> Maybe you need enough instability in your BGP, and need to hook up to
GD> a major exchange point ("something is always flapping"), or you need to
GD> do something unusual, like "change an established peer to 'no soft in'"
GD> or so.
We had some internal built of SXD1(or 2) and it worked perfectly about a year
until some moment. We don't know exactly what was done before it, but in
one day bgp just gone crazy - sometimes it started to send to external
neighbors prefixes, that definitely should be filtered by route-maps or
prefix-lists, cpu usage by bgp process gets higher and higher. So we
decided to reload box and to move to SXE2 at the same time. Till now all is
ok. So, instability could be integrated much earlier.

3 full-feeds, about 40 small peers:
#sh ip bgp sum | in network entries
196400 network entries using 22586000 bytes of memory
#sh proc mem | in BGP Router
 311   0 2592806880 1392515696  173611280          0          0 BGP Router

GD> gert

-- 
Best regards,
 Roman
 mailto:rps at cheater.ru