2 Replies Latest reply: Aug 15, 2013 1:45 PM by Wen Yu RSS

    VMware SIOC

    Mike Westerfield Newbie

      New to Nimble Storage so if this has been answered somewhere else please just point me in the right direction:

       

      I was going over Best Practices for SIOC in VMware (Using Duncan Eppings Blog post here: http://www.yellow-bricks.com/2010/10/19/storage-io-control-best-practices/)

       

      and noticed that it recommended changing the latency thresholds based on the type of storage:

       

      • Change latency threshold based on used storage media type:
        • For FC storage the recommended latency threshold is  20 – 30 MS
        • For SAS storage the recommended latency threshold is  20 – 30 MS
        • For SATA storage the recommended latency threshold is 30 – 50 MS
        • For SSD storage the recommended latency threshold is 15 – 20 MS



      I'm thinking that for Nimble storage I would follow the SSD latency thresholds, can anyone confirm this?  I don't see the SIOC setting referenced in the Nimble Storage Best Practices guide for VMware 5.



      Thanks,


        • Re: VMware SIOC
          Nick Dyer Navigator

          Hi Mike,

           

          I've very rarely seen SIOC implemented in production environments. That being said if you were to set rules on latency for SIOC, I would absolutely use Duncan's guidance on the SSD latency figures for that setting.

          • Re: VMware SIOC
            Scout

            Hi Mike, SIOC is very useful in making sure the 'noisy neighbor' does not hog up storage resources for other VMs.  As for the recommended threshold, use Duncan's recommendation as a baseline reference point.  My recommendation is to tie it to what your applications/end users demand (service level objective/agreement with application team).  20 ms is a reasonable number to set for Nimble given our flash optimization to accelerate performance.  But if your applications can tolerate higher average latency, then It'd not make much sense to start throttling back workload to get lower latency.  Hope this helps.

             

            -wen