High CPU when Next Hop goes away...

[tmarlow]

I currently have BigBand CMTS' plugged into a vlan that also has two 7200's with static routes for each netblock pointing to the various CMTS gigabit ethernet interface IPs. Pretty simple stuff, "ip route 192.168.1.0 255.255.255.0 10.10.10.1" for example. However, when I unplug the CMTS gigabit link then the 7200's cpu shoots up to mid 80's when it is normally at 5-10%. The biggest culprit seems to be IP Input process. At the same time the other CMTS's experience 40% packet loss and some serious slowness.

I'm guessing that possibly, the other CMTS' have proxy-arp turned on and there is a CEF adjacency being formed for that /32 to 10.10.10.1, and maybe the CMTS is sending back 'unreachable' messages causing the cpu to rise. Does this sound way off or are there any other hints that I could look for?

[jackyoung]

I think we need a "sh tech" at both 7200s. Are you load-sharing between two 7200s to CMTS ? How two 7200s communciate each other ? Do you mean when disconnect the 7200 A GE then the 7200 A CPU raise ? or 7200 B CPU raise ?

I suspect if you disconnect the GE and there are lots of packet in the queue, but the router is waiting to drop it all, so cause the CPU high. It is just a wild guess.

Hope this helps.

[tmarlow]

Thank you for a reply, yes, I'm load-sharing at the 7200s with hsrp, each 7200 has identical static routes to the CMTSs. 7200s run ospf and bgp, but I'm talking about when I unplug one of the CMTS connections. I did some more testing last night and my theory about one of the other CMTS' providing a proxy-arp reply is wrong, the CEF adjacency stayed in place with the correct MAC address of the CMTS, a simple "clear arp" on the routers made the CPU go down to normal and the packet loss on the remaining CMTSs go away also. I don't want to have to clear the ARP everytime we unplug a CMTS so I'm lost as to what CEF is supposed to be doing. Before I clear the arp on the router CEF shows a cached adjacency learned by arp that points to the /32 ip address of the next hop in the static route. After the clear arp command is executed it is just a glean adjacency for the attached network. I don't know if this is normal behavior or if there is something I can configure differently. I'll add the running config, but will need to change the IPs.