3

IOMeter testing says the IOPs of the storage subsystem can range from hundreds of IOPs for 100% random loads to tens of thousands IOPs for 100% sequential loads.

Is there any way I can measure or estimate the random/sequential distribution of the IO we actually have coming from a mix of servers on VMware ESX hosts?

From the SAN, I can get stats about Read Percentage, about Read cache hit rate, and about current IO counts. That says we've got ~40% read / 60% write distribution, and averaging 1,000 IOPs varying from lows of 350 to highs of 2,100.

The Read Cache hit rate is very variable, it goes <1% up to 80%, and averages 36%. Can I use that variability to conclude anything about the randomness of the IO, or is that not a good correlation?

  • 1
    It utterly depends on what SAN array/s you have, you can get a lot of information from your ESXi hosts but you've then got to collate all of that, it's far easier to look at you arrays and see if they have good stat dump options. – Chopper3 Feb 23 '16 at 17:26
  • Interpreting the Read Cache Hit Rate is going to depend on a couple of things: the amount of cache you have available on the SAN and how aggressive the SAN is reading ahead to service sequential read requests. Some of these parameters may be configurable on your SAN, so see if you can tell what they are currently set at. From the numbers you reported, it looks like you might have some significant random read I/O workloads, assuming your cache isn't too small and your read-ahead isn't too aggressive. – Thomas N Feb 23 '16 at 21:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.