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

[Gert Doering]

Hi,

On Tue, Sep 13, 2005 at 03:41:48PM -0700, Bruce Pinsky wrote:
> Gert Doering wrote:
> > I was adventurous (or maybe "stupid") enough to run 12.2(18)SXE on our
> > Sup720 based router things.
> > 
> > Yesterday BGP started acting up in *very* interesting ways ("show ip b su"
> > claimed to have seen "50" prefixes from a given peer, while the detailed
> > commands, like "show ip b neigh x.x.x.x received-routes" displayed 
> > *one* prefix, or just a hand ful).  
[..]
> > 
> > We're running various 7200s with 12.2(18)S<n>, and these never did 
> > funny BGP things - so I'm a bit confused whether 12.2(18)SXE2 has the
> > same BGP code, or maybe something completely different?
> 
> Andre Beck reported some similar prefix counting issues in 12.2(25)S on
> 7200's.  He indicated that the total networks reported in "show ip bgp sum"
> was higher than he would expect.

Indeed:

191463 network entries using 21635319 bytes of memory

on 12.2(18)SXE2, vs.

167380 network entries using 16905380 bytes of memory

on 12.2(18)S9.

> Rodney suspects CSCeh16989 which has a fix in 12.2(30)S (no 6500 or 7200
> image on that release), but I don't see a fix yet for 12.2SX.

Hmmm.  I do have a TAC case open on this, will throw the bug ID at
them, and see what will happen :-) - on a router with lots of exchange
point peers and a full table eBGP session, chances for "multiple minutes
without a single update" are very small indeed, so it might fit.


> That bug seems to have been introduced by a fix that would have been
> integrated into 12.2(18)SXE and hence you would not have seen it in
> 12.2(18)SXD, SXB, etc.

This fits to what Jon Lewis has been reporting - "no problems in SXD".

> The bug also claims to continually increase memory consumption, so you may
> want to watch that for both 1) signs that this is the problem you are
> encountering; 2) operationally impacting memory exhaustion.

The memory consumption was the reason why I initially opened the TAC
case - the BGP process is up to over 212 Mbyte of RAM:

 PID TTY  Allocated      Freed    Holding    Getbufs    Retbufs Process
   3   0 1191220080  194052656  212285264      54516          0 BGP Router      

... while the 12.2(18)S9 boxes do roughly the same stuff in 130 Mb.

> If you don't need what's in that version, you may want to consider going
> back to your previous image until this issue is resolved.

Do you think SXF would be a good idea?  Any chance of seeing this bug
fixed in SXE"3" or SFX"2" any time soon?

We're on SXE2 because we want to use BFD - which we're not actually
doing yet, but certainly want to enable "real soon".

thanks a lot for your help,

gert

-- 
USENET is *not* the non-clickable part of WWW!
                                                           //www.muc.de/~gert/
Gert Doering - Munich, Germany                             gert at greenie.muc.de
fax: +49-89-35655025                        gert at net.informatik.tu-muenchen.de