April 7, 2009 - Sven Huisman
Citrix XenApp on VMware ESX: 1 or 2 vCPU part 2
One month ago I wrote a blogpost about using 1 or 2 vCPUs on Citrix XenApp VMs. My conclusions in that blogpost were:
– 2 vCPU terminal server (or XenApp) VMs allows more users and gives a more consistent user experience
– Use dedicated hardware (ESX cluster) for the TS/XenApp VMs
– Don’t overcommit total amount of vCPUs
I stated this conclusion to get some reactions and it worked. Because as Jason Boche commented on the article: It depends!
“It’s all about the workloads and each company and environment will have different workloads. Traditionally speaking, published desktops chew up a lot more server resources than published applications.”
I couldn’t agree more, it depends on the workloads.
For the first point, Duncan Epping from yellow-bricks tested in a lab-environment and he stated that he could get more users on a box with 1 vCPU machines than with 2 vCPU machines. That could be true, because Alan Renouf tested it and came to the same conclusion, but how about user experience?
Recently I’ve tested both 1 and 2 vCPU VMs with Citrix XenApp in 2 different (but similar) environments. Both environments uses published desktops with RES Powerfuse for user environment management. Hardware was similar as well: a dual quadcore server with 24GB RAM.
Scenario 1 had 5-10 users on each VM and in scenario 2 each VM had 20-25 users on it. In these 2 cases, 2 vCPU VM’s gave a better user experience than a 1 vCPU VM, regardless of the number of users on each VM. In this case a better user experience can be a faster login time, faster application startup times, less screen delays, etc.
Also, I monitored the %RDY time on the VMs. When I did not overcommit the number of vCPUs (except for the use of service console) the %RDY time was good: between 0% and 4%. That was with 4 dual vCPU XenApp VMs. When I added a 2 vCPU XenApp VM, the %RDY time started to spike: 5% – 35%. Overcommit on CPU is a NOGO with 2 vCPU (XenApp) VMs.
So XenApp in a virtualized environment can be tricky. In the next project where I’m going to use published desktops with RES Powerfuse, I probably go for 2 vCPU XenApp VM’s, but it depends… ;-). At least I now know that using 2 vCPU on a virtual XenApp server isn’t a NOGO anymore.