Hi all,
I’ve started working at a new place and need some guidance going forward.
I’ve worked with XenApp and VMware before but it was not this advanced.
Current setup:
Dell R720
GRID K2
ESXi 5.5
XenApp 6.5
PVS 6.1
Users have thin clients and connect to a shared Desktop using the Receiver.
The Citrix farm is setup with a bunch of PVS’ed XenApp VM’s. Each VM has assigned one GPU via PCI passthrough (or whats-it-called?!).
This setup is not ideal; VMware HA is not working on VM’s with GPU passthrough enabled, the VM console in vCenter is black with a black dot as the cursor (RDP works) on each machine due to the GPU (not sure what’s going on). And, I’m not even 100% sure that the GPU passthrough is operational and working for the users.
Anyway.
I see that ESXi 6 lists support for vGPU. And that you can, instead of letting one VM hog a GPU all for itself, share the GPU’s over many VM’s.
But I can not find any information if that is possible to do in vSphere (I can only find info about VMware Horizon) and/or with XenApp.
What would you do in my shoes? I am planning for major upgrades after the summer vacation, not sure what I should upgrade though :)
Sorry for any crappy english :)
Regards
P-H, Sweden
Citrix supports XenApp with vGPU and implementations of NVIDIA GRID vGPU on both XenServer and vSphere deliver the same functionality.
So yes you can deliver XenApp on vSphere with the XenApp VM running vGPU, and it is supported.
However, the question is, should you use vGPU with XenApp?
Depending on the load on the GPU from the users on the XenApp server, they may be better served by a single GPU in passthrough.
If you have a low number of users for a particular application then vGPU may make sense. But if you’re looking at larger numbers, and potentially with higher GPU demands then usnig vGPU with XenApp may be counter productive and you could find yourself with an increased number of XenApp servers each with a small number of users on them so increasing your administrative load.
It’s not something with a simple yes/no answer. It’s fully supported, but may not be the best choice.
I agree fully with Jason. The best way to find out what works best is to try it both ways. A single XenApp instance with 8 or even 16 VCPUs can do an awful lot of processing when teamed with a K2. With fewer users active, each will also get more of the capabilities of the GRID as opposed to being always limited to a mximum slice of it in vGPU mode. It’s also critical to allocate enough VCPUs and memory to the clients, so monitor your loads carefully when deciding on a configuration standard. I’d also strongly encourage upgrading to XenApp 7.6 FP2, which now supports Framehawk, and improved performance.