ESX U3 performance for VDI

[Cameron2007]

I am looking at a VDI deployment and want to know what peoples experience is. I have a couple of solutions in mind using various hardware. HP blades BL680's, DL585's and DL385's G5. The initial designs were based on either 680's or 585's (4 sockets each) however due to costs I want to look at implementing a cheaper hardware platform. This leads me to the 385 solution (2 sockets) but due to the amount of VM's reccomended per core I am looking for some re-assurance. I am assuming that the memory across all platforms will be the same.

Previously the rec per core was 8-10 VM's however I have read that ESX U3 has support for up to 20 per core. Has anybody implemented a solution based on 20 VM's per core in a production environment? And what are everyones thoughts on this improved functionality in U3?

Thanks in advance

[szekelyk]

You should read this technical paper: http://www.vmware.com/files/pdf/vdesk_scaling.pdf

To cut it short, the conclusion of this paper 7 View VM per core.

But YMMV!

I mean for this specified workload the result is 7VM/core, butt different workload means different VM/core.

[Cameron2007]

I had scaled at 7 per core initially as I was aware of the scaling. However this document refers to Update 2 and the vmware website states up to 20 on Update 3 in the what's new section.

http://www.vmware.com/support/vi3/doc/vi3_esx35u3_rel_notes.html#whatsnew

So still not sure what the situation is in a production environment.

[TomHowarth]

20 vm per core is posible but only is task based environments, on normal working environments I have obtained 10 -15 (YMMV) , that said that was on BL685 with 16 available cores. you will only have 8 cores and as such the availability of free cores to service CPU instructions will be lower. I would feel more confortable with the lower end of my second estimate on dual/quad cores

Memory will be sorted out be TPS and you will recieve significant saving there. your next decision point is traditional VDI desktops or Linked Clones. here the moot point is Storage costs. also remember to disable the automatic Defrag if using linked clones with Vista.

If you found this or any other answer useful please consider the use of the Helpful or correct buttons to award points

Tom Howarth

VMware Communities User Moderator

Blog: www.planetvm.net

[Cameron2007]

Hi Tom,

the department we are rolling out will initially be "light users" and due to the structure of this department it is unlikely that we will ever reach the maximum concurrent connections that the solution requires. We will probably use linked clones and will look to utilise Netapp de-dupe on the storage so storage costs shouldn't be a concern.

If we don't get the appropriate performance on the initial number of nodes the temptation would be to scale out (more nodes) as opposed to more CPU and RAM per node.

Thanks for your input

[szekelyk]

AFAIK, 20 VM/core is the maximum supported amount in ESX/ESXi , the 7 VM/core is measured amount in View enviroment under specific workload, on the specified hardware.

[TomHowarth · Accepted solution]

building out is most likely the better option, but remember to try and buy new Hosts in at least pairs, for vmotion compatibility, however with the increace in EVC enabled Chips that is becoming less of an issue.

If you found this or any other answer useful please consider the use of the Helpful or correct buttons to award points

Tom Howarth

VMware Communities User Moderator

Blog: www.planetvm.net

[Cameron2007]

Thanks Tom that puts my mind at rest a bit. And I will buy identical boxes (i've been in CPUid masking land before- never again )