I am trying to do a POC for an app in XenApp. I have a C240 with 2 K2 cards and I can run 32 VDI’s per server with no problem. We would like to give it a try to XenApp to see how density increases (or decreases). I was able to install a 2012 R2 server, XA7.6 with full GPU passthru on XenServer. I see that I can install a maximum of 4 XA servers per host. How many users I can run simultaneously on each XA and where I can see/setup this? Thank you
From what I know it really depends on how much GPU the app uses for each user connected, together with CPU, memory etc. All users on each XA server will be sharing that one GPU core. Limiting the amount of users is done in the built in load balancer in XenApp.
I don’t think there is an answer to your question; other than test and see.
Thank you - I was "afraid" of that. I will get HPLR setup so I can do some testing.
See also: https://gridforums.nvidia.com/default/topic/450/xenapp-with-nvidia-grid/configuring-multiple-grid-engines-on-a-single-xenapp-virtual-machine-running-on-xenserver/
While not officially supported, it would allow you to see where your XenApp server runs out of steam while loading more and more apps that tap into the GRID. In addition, the splitting up of resources won’t be as controlled by whatever happens to be run on one XA instance, which might be over-utilizing one GRID engine while the other is barely used. In my experiments, the GPU load seems to split up pretty well as the number of running applications was increased.
That aside, I agree that there is no real way to tell what setup is "better" other than to try different options out in your won environment. The same can be said for using a dedicated vGPU per VM vs. GPU passthrough.
Citrix recently released an updated statement about experimental support of multiple GPU’s in XenApp.