HPE Storage Users Group
https://3parug.com/

Use Iometer to test performance
https://3parug.com/viewtopic.php?f=18&t=2817		 Page 1 of 1
Author:  marius_roma [ Fri Mar 09, 2018 9:54 am ]
Post subject:  Use Iometer to test performance
Let me introduce myself, I just added a 3PAR 8200 to an existing VMware vSphere infrastructure.
I need to measure 3PAR performance compared to other existing storage.
So, I created a VM and installed Iometer.
I moved the VM on a datastore hosted by the 3PAR and made some tests with Iometer.
Then I moved the VM to another datastore hosted by a different storage and performed the same tests.
Let me ask expert people:
  • Am I using the right tool to compare performance?
  • How should I confngure Iometer to get the more meaningful results?
  • Is there any sample configuration file I can start from?
  • Should I do anything different, given that either the infrastructure and the 3PAR host production VMs and I cannot create problems?
Regards
marius
Author:  Richard Siemers [ Fri Mar 09, 2018 10:32 am ]
Post subject:  Re: Use Iometer to test performance
IOmeter is not a good choice for benchmarking modern arrays because it uses easily deduped data over and over again for writes. IOmeter is easily tricked by advanced array features like thin provisioning and dedupe. Ditch it asap, it hasn't been updated in what, 10 years?

The tool I recommend is VDbench. Its free form oracle. You will have to create a free account on their site to download it.

It takes time, physical resources and good knowledge to do a meaning benchmark. Benchmarking on production arrays, or production ESX hosts is not free from risk. If you do manage to perfect your benchmark plan, then either the array or the ESX hosts will be at 100% util and impact production. If you don't drive it hard enough to cause a production impact, then you are not doing enough to benchmark the system. Its a catch 22.

Parallelism is key. Last benchmark I did used 4 ESX hosts, with 1 VM each. All were tuned with vendor best practices, much of which involved increasing the queue depth of the Guest OS, the Guest virtual machine hardware, and the physical ESX host hardware settings to open the pipes as wide as they could go. There is only so much a single hypervisor IO scheduler will allow and that needs to be leveraged against the capabilities of the array to be benchmarked. You have to size the benchmark platform to meet your requirements.
Page 1 of 1 All times are UTC - 5 hours
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/