2 Replies Latest reply on May 19, 2014 10:04 AM by bbeut



      To monitor virtual machines I want to see the CPU average and peaks value of a virtual machine.

      For the peak CPU value I am using the query: vm.cpuloadSumpeak.month (%).


      However, many virutal machines have a peak above 100% - some are even 600%.



      Can somebody explain to me how the peaks are being calculated and/or how I should interpret the peaks?



        • Re: HIGH CPU PEAKS



          could be the reason that the VMs with CPU load higher than 100% have more CPUs? Something like this.


          If this is not your case please contact support.

          • Re: HIGH CPU PEAKS


            I had this exact same problem and opened a case with SW on it. Over the course of 3+ weeks (I wish that were an overstatement) I was finally told that the cpuloadSumpeak was a broken metric and to be removed 'at some point in a future release'.


            Here is the exact quote from a VMAN Sr Product Manager, from 4/23/14


            You had a completely valid point on the VM CPU Used % metric. After discussing the counter (CPU Used %) in question, we determined that the algorithm VMan is using is not correct as it tries to turn a delta in milliseconds of CPU time into a %, but with an uncertain denominator (also in ms). The CPU Used % counter has some historical context for why it’s there, but that is no longer needed. After reviewing the case in detail, it appears that this is still correct except in some corner cases, but not perfect and thus will be removed in the future.


            Luckily, the “fix” is already in the product “CPU Utilization of Host %” (performance statistic) or “CPU Load” (configuration statistic) are counters that VMware now provides natively and are collected by VMan and these should be used for your alerts, reports, etc. instead of CPU Used % in the meantime. We will modify the defaults in a future release to use these counters for VM CPU Usage % on VMware.  Unfortunately, we are not allowed to provide timeframes on features or fixes, but you should not have to wait at all to start using the “correct” counters in your deployment.


            I hope that helps at least answer your question - but not resolve the problem. I too would LOVE to have an easy way to report on peak utilization over a period of time, as right now VMAN loses all peaks for just about every metric due to the way their graphs show normalized data. Very difficult to show a customer their CPU usage on servers over a month when it normalizes out in 12-hour increments.