Hi everybody !
I'm trying to use IOmeter to test a new AFF220 metrocluster.
However, i don't really know which parameters to use to test it...
How many workers should i use ?
I don't really know what is Tranfert Request Size, Align I/Os if you have any ideas ?
I did a try with 4K 100% random on a windows VM with 10 workers. Display results are showing 36K IOPS, however, on the LUN where resides the VM, can't go over 2000 IOPS.. Don't get that !
Synthetic IOPS Testing is not.. a great measure of anything. Have a view of this video for a bit of a discussion about why - https://www.youtube.com/watch?v=Gfg1pNMCqUQ
Are you concerned by the performance of your system? or rather, what brings you to want to test?
Thanks for your answer.
I just wanted to guarantee that the metrocluster supports for example 30K IOPs. However, we don't have that much VMs on it so wanted to use a tool to "stress" the metrocluster to see how it goes.
Here we come to the problem of IO speed tests.
As you say, you have run one test which shows 36,000 IOPS, therefore, it can do 30,000+ IOPS.
What is the difference between the networking, connection etc of that test and the one which shows 4,000 IOPS?
Yes i can see 36K IOPs from the IOmeter application.
However, when i had a look on the LUN where we simulated IOPs, i can't go over 2K IOPs.
Wanted to understand what could explain this difference... I tried to had a look as well on the volume or aggregate and same problem.