Nov 26, 2012
We installed a solution with 2 Cisco 2801, BGP multihomed failover.
View 8 Replies
1) The router which is currently getting all the traffic gets to 55% to 60% of CPU usage when handling 40 SIP/RTP streams . This equals 10Mbit up/10Mbit down and it showed around 5800 packets TX and around 5800 packets RX, with a majority of them CEF switched. As those figures are way less than the performance figures published by Cisco, we wonder if we made any mistake in setting up our router, or if we can do something to improve the router setup.
2) Does it have an impact on router performance if we increase/decrease RTP packet size, thus increasing or decreasing the pps relative to the consumed bandwidth?
3) If it is not possible to improve router configuration, we also wonder about possible replacement units for those routers. Would a 2901 do a good job? By how much would it rise the capacity? What other models would you recommend if we plan to rise the number of concurrent calls by a factor of 4 or even 8 times of what we have now (so up to 48000 pps and 80Mbit).
Here is what we tried:
- ip route-cache same-interface does not seem to improve anything
- ip flow ingress on or off makes no difference
- disabling the inbound ACL on fa0/0 seems to reduce load by 10%, although I don't understand why - a very high percentage is CPU interrupts, and ACLs are process switched, or not?
- we tried following the Cisco guide for high CPU due to high interrupts, with no success
Here are some usage statistics:
The graphs that we plot via SNMP show a propotional growth/increase of CPU and bandwidth (and thus pps) At the highest loads, we had a bit more than 55% CPU utilization with more than 50% interrupt CPU.
CPU utilization for five seconds: 36%/30%; one minute: 30%; five minutes: 30%
PID Runtime(ms) Invoked uSecs 5Sec 1Min 5Min TTY Process
127 13140 954 13773 2.00% 0.29% 0.07% 194 SSH Process