vCPUs and Citrix PS 4

[peterhigsonuk]

I've currently implemented a (rebuilt from scratch) Citrix PS4 farm instance, using 1xvcpu's in each of 25 hosts,

I'm doing some capacity testing, and seeing a slower successful login rate than with my physical farm, and looking at my perfmon dumps, looks like the only resource contention is CPU,

what are anyone's experiences/practices regarding multiple vCPU configurations for Citrix PS4 Servers, has anyone seen any core contention issues or performance degrading when using more vCPUs? I'm planning on adding an extra vCPU to each citrix server, leading to a total of 50 VCPUs for the farm.

(A little info regarding the servers, they are win2k3r2sp2 standard boxes, 4GB ram, 4GB reservation and no vswap, Citrix MFPS4, running approximately 40 sessions/box), the hardware is 5 blades 4x4core Xeon bl680's)

comments regarding any prior experience with such configs would be appreciated

[depping]

Project VRC did a performance test with 1 vCPU and 2 vCPU vm's and the 2vCPU's clearly outperformance the 1 vCPU vm's. Although I haven't been able to confirm it it does make sense...

http://www.virtualrealitycheck.net/

Duncan

VMware Communities User Moderator

-

---

Blogging: http://www.yellow-bricks.com

Twitter:

If you find this information useful, please award points for "correct" or "helpful".

[kjb007]

PS servers with 1 vCPU is generally the best practice, but your mileage will vary. This wil depend on what type of app you are running as well. The real power of Citrix on vmware is the scalabiltiy and number of instances. How did you determine CPU contention?

I would typically use 1 vCPU vm's, but I've used 2 as well.

-KjB

VMware vExpert

[M__Y_]

Hi,

I had same problem on ESX 3.01. I have stopped VMware tools service and change the display hardware acceleration (Control Panel/Display/Advanced/Troubleshoot) to None .

Regards.

[CCJNL]

These three blogs are great reads on this subject. The last one is quite interesting and speaks to your question.

http://virtualfuture.info/2008/07/citrix-on-vi3x-recommendations/

http://virtualfuture.info/2008/10/more-xenapp-45-on-vmware-recommendations/

http://virtualfuture.info/2009/03/citrix-xenapp-on-vmware-esx-1-or-2-vcpu/

[peterhigsonuk]

Thanks guys, seems I get more logins with 2xVCPU machines at the net loss of some raw processing power on the cluster from the tests and studies I've made, i've been watching the CPU Processing queues, and it seems actually from some real world tests that the applications run faster on my new VMs than the old physical instances anyway,

I've mostly been looking into the Load evaluator in PS4 now to tweak how it deals with the resource usage, and I think i've come to a good balance for our implementation by punching up the Page swaps max out value to 500, and uping the Context switches/sec to 16000, and dropping the CPU metric entirely, I managed to launch 120 applications on one VM, and that's more than enough than required,

Has anyone any further experience with balancing load evaluators/VM characteristics?

[peterhigsonuk]

We had improvement here by adding 2xVCPUs to our guests, and increasing the number of Citrix instances to cover a percieved scale-out deficit. this seems to generally have improved things,

A highly tweaked load evaluator seems to also ensure a more steady balance on the citrix farm, and this is definitely something people should look at closely, especially with similar PS4/ESX configs as ourselves

[peterhigsonuk]

Ta duncan, this is the most significant benefit to us here.

[msemon1]

We started with 1 vCPU and upgraded to 2 vCPU's. Seem to get a little better performance. Take a look at these best practices for Citrix on Vmware

http://virtrix.blogspot.com/2007/03/vmware-best-practices-for-deploying.html

http://viops.vmware.com/home/message/1573#1573

One thing to look at is how you setup VMware tools if you ar doing folder redirection. We excluded installed shared folders.

Mike