0 Replies Latest reply on May 24, 2012 5:05 PM by tomaddox

    IOPS vs. Peak IOPS

    tomaddox

      I'm trying to understand the difference between IOPS and Peak IOPS in SWVM. My understanding of these metrics, which I believe to be faulty, is that the IOPS metric is an aggregate of the stats over the collected interval, so vm.iops.latest, for example, would have the latest number of i/o operations reported by a VM, vm.iops.hour would be a rollup of the total iops over the past hour, vm.iops.day would be the rollup of the past day, etc. By contrast, vm.iopsPeak.latest would be the latest high-water mark reported by the VM, vm.iopsPeak.hour would be the highest IOPS reported during any collection over the last hour, and vm.iopsPeak.day would be the highest IOPS reported during any collection in the past day.

       

      Is my understanding correct? It appears not to be because I am seeing a VM reporting iopsPeak.hour of 1792/s, but when I look at the performance chart, I "only" see a peak of just over 200/s in that timeframe, and when I add the IOPs Peak counter, a message is displayed indicating "No data available for this time period." It seems like this issue is related to the VM in question being on an NFS datastore, so perhaps the iopsPeak counter is not valid for NFS datastores.

       

      The other thing that's confusing about this is that SWVM shows statistics that do not seem at all related to what's being reported by VMware. For example, I'm looking at the output from a particular VM, and it shows 428 IOPS most recently. Looking at the same data in vCenter, the highest number of disk requests I can find in the past hour for that VM is 63.

       

      Can anyone help clear this up for me?

       

      Thanks,

       

      Tom