09-17-2009 12:36 AM - edited 03-06-2019 07:45 AM
Hi,
Is there any way to see what is causing the load on the cpu?
We are running a 6504 with sup720 and 12.2(18) sxf16.
The show proc cpu gives me an average of 60 - 80%, but the details only reveals 5-10. How can I see what traffic is actually causing the load?
09-17-2009 12:40 AM
Hi
You can do a show proc cpu sorted to sort according to order (Highest load process on top).
You can then see the process name that is hogging CPU resources. Post the process name here if you need more help.
Most of the time it is the logging daemon hogging CPU (because of sudden increase of logs)
09-17-2009 12:46 AM
Yes, and add | excl 0.00 so I only get the ones actually having something other than 0.
But it only shows a few % of the total load.
Can I somehow see the rest?
09-17-2009 02:01 AM
It may not be process utilisation moreso interrupt utilisation
Could you post the output of
show proc cpu sorted 1min
Also clear all your counters and do
show int | i output drops
09-22-2009 10:19 PM
I solved it by setting up CoPP.
That way I could find a few servers with the wrong MTU.
Changing the MTU on the servers solved our problem.
09-22-2009 10:34 PM
Hello Ola,
you have been kind to provide a feedback on this issue.
This helps to improve the forums.
CoPP by rate limiting packets sent to main CPU had helped and you have found an explanation of the problem.
Best Regards
Giuseppe
09-22-2009 10:37 PM
Hi,
Can you let me know more details on "CoPP"; i am also facing similar issue like high utilization, it would be helpful to know more on this.
Regards,
Amol More.
09-22-2009 11:33 PM
Hi,
Sure, it is a QoS feature for traffic to the cpu, with a bonus of DOS protection.
You set it up with access-lists, class-maps and policy-maps, then bind it to the control plane with a service-policy.
The way I did it was to set it up without dropping any traffic that exceeds or violates the policy but transmit all so I wouldn't drop any traffic accidentally. You still get hits in the access-lists so that you can see where the traffic is originating from. The service policy is not needed to get hits in the access-lists, but it would give you a bonus of finding out how much traffic you have for routing, arp and so on. Make sure to add an entry for permit ip any any at the end of the access-list.
You can capture non ip traffic by matching the class-map to a protocol instead of an access-list.
Check the configuraqtion guide for your version of the IOS.
Then check the hits in the access-lists and the traffic rates with show access-list and show policy-map control-plane input.
Hope that will help you out.
/Ola
09-23-2009 12:11 AM
Oh, I forgot, once I figured out what ip addresses was generating the traffic i used the NAM to capture it and find out why the traffic was sent to the cpu.
But any sniffer would work for that.
/Ola
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide