3 Replies Latest reply on Jul 9, 2009 11:11 AM by bshopp

    Feature change: CPU utilization calculation

      Regarding case 104360 that I opened on the NPM reporting of CPU utilization on a Cisco 6509,

      “The CPU Load data does not reflect what I see directly on the Cisco 6509 CLI. NPM indicates a 95th percentile load of 2% when polling every 5 minutes while show proc cpu reports 8%.”


      In response, this link was provided along with an explanation of how the utilization data points where acquired. https://www.cisco.com/en/US/tech/tk648/tk362/technologies_tech_note09186a0080094a94.shtml


      The Cisco document is not just suggesting that the results from OID or

       be averaged, but in continuation that follow on OIDs be requested and that result used to delineate between cpu utilization for various functions. This is defined in the document under “Procedure for Devices with Multiple CPUs”.


      As an example, the individual line cards on a 6509 are performing independent functions. It is of little value to have it reported and graphed that the average cpu utilization of the 9 slots is 10% when that could mean each and every slot is running at 10% or that 1 slot, the Supervisor, is running at 82% and the rest are at 1% each.


      Please review and redefine CPU utilization. The ultimate goal should be the ability to select utilization by module.



        • Re: Feature change: CPU utilization calculation

          I will log this as a request into the system.  You are correct in your statement, today we take the output and average them for the chassis.  In  the meantime, if you want, you can create UnDP's to display the results of each processor in tables and view the percentage of each processor.

          So what you would like to see is a table for each processor on the box with the appropriate label as well as allow you to choose which processor you would like to trend? Is the average giving you the general state of the box useful to you?

            • Re: Feature change: CPU utilization calculation

              Answer to first question is yes, that would give the most flexability. Answer to second question is no, certainly not in this particular case. Could be a different answer for other chassis but in the 6509 case the CPUs per slot are performing such different functions that an average across the box doesn't mean much. The line cards are switching, the Sup cards are redundant so only one is active, the NAM is used for trace collection having nothing to do with packet forwarding. So averaging them all together just delutes the true utilization of each funtion. Thanks for your interest.