[c-nsp] cef load-sharing per-packet bizzare behavior

[Andy Dills]

We just moved a 2xT1 customer over to a 7513 as part of some network
consolidation.

We're doing per-packet CEF for this customer. (We've done per-packet CEF
for half a decade now, so I'm well aware of the simple gotchas.)

I've confirmed per-packet CEF is enabled on the interfaces, as well as
dCEF.

I've even turned on 'ip cef accounting load-balance-hash', and watched the
output of 'sh ip cef <their loopback IP> internal", and watched the packet
hashes increase pretty evenly. From what I understand of debugging CEF,
this indicates that at the CEF level, they are being switched properly.

Yet, when I watch the txload on the interfaces, one of the T1s seems to
get almost exactly double the txload of the other T1 interface. The
relationship between these numbers seems too coincidental to me.

Running 12.2(16).

Does anybody have any ideas or suggestions?

Thanks,
Andy

---
Andy Dills
Xecunet, Inc.
www.xecu.net
301-682-9972
---