Friday, 22 May 2009

ESX IOPs - This is NOT a HyperV bash!

Now I'm not a nasty horrible person (although some people might think that) I just like proving facts, a google to find material on Virtualising Microsoft Biztalk and I stumbled across this excellently written Microsoft paper on how BTS benchmarks when running on Hyper-V against conventional physical hardware.
Conceptually this document gave me some great resource material to use as I have no idea on how Biztalk server works, how the architecture of Biztalk is configured on the box or how components figure across a landscape. The best bit for me was that Microsoft kindly provided the Storage IOPs profile to compare and benchmark, this gave me the following quoted figures for expected IOPs when virtualising Biztalk on Hyper-V;
Now amateur alert here......I currently only have a very basic test and development platform rig currently running ESX 3.5, its only got DAS and is running a RAID 5 across about a zillion disks so is not the best system to provide me with the 400000 IOPs that is touted as being possible with vSPhere. The specifications of the Hyper-V VM against my sh*tpit server were as follows;

To gain a quick comparative idea on how a Microsoft quoted Biztalk simulated workload would run on an ESX setup I set to test the benchmark IOmeter parameters within my ESX 3.5 VM to see what comparable IOPs and Other readings I would achieve by using VMware ESX.

The results I experienced were as follows;

I think the results speak for themselves even with ESX being run on a lower expected performing platform. Now please don't sue me Microsoft for writing about this....I really give you upmost credit for writing this document as it gives someone the opportunity to learn how your application stack works and is expected to perform and would like to see more material and vendors following suit (including VMware)

Dan, You compared to completely different disk subsystems. SAN vs DAS as well as two different RAID stripes. This isn't even close to an apples vs. apples comparison. Kind of like an apples to Buick comparison.

I don't understand what makes you think that this was an Apples for Apples comparison, i'd say it just highlights the obvious about ESX performance on a lame underoptimised system.

Are you basically saying an ESX VM will show lower IOmeter results when it is run on a similar highly optimised storage configuration as per used to obtain the published results for HyperV?
@Dan, that's really funny... so on your "sh1tpit server" you get better results than an optimised Hyper-V configuration? Can you add a column for the physical results, perhaps include some specs for them? Or should I stop being lazy and follow the link back the M$ site... ;-)
Can you reverse the comparison servers and test again? That is put Hyper-V on the 580 with the DAS and the ESX on the SAN with the other variables the same as now.

Post a Comment

Subscribe to Post Comments [Atom]

<< Home

This page is powered by Blogger. Isn't yours?

Subscribe to Posts [Atom]