[c-nsp] IP Input high CPU utilization

[David Coulson]

David Coulson wrote:
> So, I decided to make core1 the active router in HSRP. It started to
> utilize 100% cpu on core1 due to the 'IP Input' process. I know this is
> caused by process switched packets, however since the config for the
> interfaces is the same on both routers, I can't figure out why it would
> behave differently on the two routers. For all intents and purposes,
> these two routers are identical (same IOS, same RSPs, same VIP/PAs).
> Even when I have HSRP move onto core2, we still have outbound traffic to
> that interface on core1, and traffic which would previously have been
> routed via core1 (e.g. directly connected customers) will come in on
> another GigE interface sub-interface and end up on core1 anyway.

After some digging around, I was able to fix this issue, although I'm 
not entirely sure how I fixed it... Why it would be different on the 
GigE Vs. FastEthernet is anyone's guess, although Rodney's suggestion 
that CEF is distributed across sub-interfaces may contribute to it.

Basically there were a bunch of routes on core1 that caused a routing 
loop, and of course, TTL <=1 traffic is punted to the processor. Once I 
removed these routes, CPU usage dropped significantly and I was able to 
bring up the GigE sub interface under HSRP without any issues. I've been 
consistently under 8% CPU utilization (process + interrupt) all day, 
which is much better than we've seen in the past.

David