[c-nsp] Funny (and hard to describe) AWOL routes

[Rick Kunkel]

Hello all,

I've got a problem that I unfortunately don't know a heck of a lot about, 
which I understand makes answering this question difficult.  But I thought 
I'd pick people's brains in the hopes of at least getting pointed in the 
right direction.

Here's the deal.  I run EIGRP for the internal network, and BGP to talk to 
the outside world.  Occasionally, I go to add a new block or router to the 
EIGRP network, and it just won't work.  Strangely, all the "show ip route" 
commands look good, but traffic just won't get where it's supposed to go. 
Crazily, sometimes I can get 20-40% of packets through to the internet, 
but traffic to the internal network is usually lost.  However, it 
sometimes seems as if the traffic might be lost at a border router we 
have, which is currently getting two full route tables on a Sup2 running 
IOS 12.1(26)E8.  (Yes, I know, impossible.)  In an effort to minimize 
downtime, I can only poke around at things for so long, before performing 
the wonky fix.

The fix...  Usually I can do something that will withdraw or otherwise 
change the new announcement, and then put it back, and it will work.

The LAST time this happened, however, when I re-added the new block, 
suddenly another block on our network became unreachable.  It was as if 
the new block kicked the old one out.

To me, this smacks of a memory shortage somewhere, and it's occurred to me 
that it may be that border router that has a bunch of EIGRP stuff AND the 
BGP stuff.  I've heard tell of the TCAM filling, but that's supposed to 
log messages, and I've seen none of those.  And does EIGRP use the TCAM? 
Perhaps an OS bug?

Anyhow, I don't expect anyone to solve the mystery for me (unless they 
immediately know what's causing it), but I was hoping for some direction. 
Any commands I can run to quickly show me exhausted space, etc.?  I know 
the Sup2 needs to be upgraded, but I find myself wondering if that will 
fix the problem, or if this is a result of something else entirely.

This is not a horrendous problem, but it rears its head from time to time, 
and makes things difficult.  Most frustrating of all is that I can't get a 
bead on it, and have minimal time to troubleshoot on a production network.

Any ideas and/or pointers?

Thanks much!

Rick Kunkel