Benchmarking Windows VMs

[Magister75]

I have two iSCSI LUNs, one configured as RAID 0 and one configured as RAID 5.

When my VMware cluster was set up,the vendor recommended the RAID 0 configuration for better performance.

I have done some basic tests, and can see no significant difference between the two.

What is recommended to thoroughly test this?

All I have done so far is create a VM on each and repeately write then read then delete a file. As I said, no significant difference from the VM's point of view.

Would there be significant difference from the point of view of the SAN? How could I tell?

If there is little difference, I will reconfigure as RAID 5 and be happy with the extra space.

Otherwise, I'll reconfigure all as the performance optimum.

Thanks,

--MB

[azn2kew]

Have you look at IOmeter tool to test your I/O performance? What types of disk do you have and how many? There are definitely performance impacts for RAID 0, 1, 5, 10 and 50...and the higher the more overhead you've wasted so depends how many disks and types of application you want to run and choose the appropriate one. For the most part RAID 5 would do just fine, but for high I/O such as Exchange, SQL, Oracle RAID 10 or 50 would be great but remember overhead.

If you found this information useful, please consider awarding points for "Correct" or "Helpful". Thanks!!!

Regards,

Stefan Nguyen

VMware vExpert 2009

iGeek Systems Inc.

VMware, Citrix, Microsoft Consultant

[LucasAlbers]

I have to agree with him, you need to test with iometer using a test that simulates your expected utilization.

The different raid levels performance varies depending on the type of io load.

Example

Raid 5 works ok for a dates transaction log, sequential fixed writes, but worse for the database files.

[Magister75]

Thanks for the information on IOMeter, and on the information about Raid 5 and database files.

It looks like IOMeter is going to take some time to figure out.

I'll work on it

--MB

[Magister75]

I will follow up the suggestions given.

[LucasAlbers]

The problem with raid 0 is the lack of redundancy, I would never pick this for production data.

For testing yes.

For real vm's or data, never.

Your san should have so many spindles that you should get good enough performance with raid 5 or raid 10.

It all comes down to what type of vm's you will be running and what their load will be.

The advantage of virtualization is that you can just present your shared volume for use by all your vm, this assumes you get adequate performance.

Vmware has a number of whitepapers that compare disk performance.

Vendors also have reccomendations for their application.

My default deployment is raid 10, it has better performance and reliability, albeit I get less space.

The 2nd choice is raid 5 for vms' that need lesser performance and or more space.

You can spend a long time doing benchmarking, when in practice a good default will probable be adequate for your needs.

Their are huge threads on comparing performance with iometer, tons of differing views and information.

[Magister75]

Lucas,

I can't believe I typed that..

My choices in this situation are Raid 5 and Raid 1. (mirrored)

I cannot configure Logical Volumes on this SAN as Raid 10.

Thanks for your input.