0

When testing IOmeter I know always need to be avoid hitting the cache if the logical drive with the OS is testing, some I have some question about it:

  1. Do I need to avoid the memory on the controller or the machine? Let's say I have a controller with 16GB of Cache Memory, creating a file of 20GB will be fine?

  2. When testing Raw Devices will IOmeter avoid the cache automatically?

  3. Why avoid Cache Memory if its will help a lot for the IOPS generation? Aren't all the storage vendors will show the similar results.

Thanks in advance.

Juan Mulford
  • 3
  • 5

1 Answers1

1

1.You should size your workset to match the expected cache hit ratio. This totally depends on the workloads you are expecting.

When you want to get a nice high number make sure that the workset completely fits into the cache.

If you want the number for the drive which is like if there was no cache at all use the maximum possible size for the workset.

For everthing in between you have to estimate the cache hitrate of your workload.

  1. IOMeter does not avoid the cache for raw devices.

  2. Vendors use scenarios for their benchmarks which contain up to 99% hitrate. So for daily usage these values are absolutely worthless.

cl_lime
  • 26
  • 3