Bad disk latency for VM client

[danswartz]

Just talked to someone who's company tried replacing their exchange server with a VM hosted on an ESXi box. Exchange performance sucked due to "disk latency". He asked a few questions and found out the datastore was on a local SATA disk, so they moved back to the physical server. Hopefully this time they will have someone who knows what he is doing consulted. Sigh...

 

[AMD_Gamer]

What else were they doing?

 

[danswartz]

What else were they doing?

but there are several hundred exchange users on that box. not hard to imagine swamping a single SATA drive...

 

[NetJunkie]

Bad storage design is the #1 cause of bad performance in virtualized environments. People make bad assumptions.

 

[danswartz]

Heh, just talked to my buddy. Apparently, there were 'free cycles' on that host, so they threw a windows server 2008 VM on it, gave it N gigabytes on the local datastore and said 'go for it'. Live and learn...

 

[AMD_Gamer]

what else do they have running?

 

[danswartz]

That I can't say, specifically. I know there is non-trivial disk load already on that box. On a single SATA drive. Can't say I'm surprised at the result.

 

[lopoetve]

what else do they have running?

Doesn't matter. Exchange will use WAY more IOPS than single, or even 3-4 SATA disks could possibly provide, especially with more than a couple of users.

 

[bearlevel]

There are performance matrix out there that will tell you the # of spindles and IOPS you will need per mail box in an exchange environment. To even attempt this on a single disk was just a big waste of time.

 

[NetJunkie]

Yep. There is a good sizing spreadsheet for that.