I’m designing a XenApp 7.13/Win2016 environment on vSphere 6.5 with Tesla M10 for 500 users. All users are office workers (common tasks like Office, webbrowsing with video content etc.). I‘m planning virtual apps as ccu licensing.
My big problem is to understand the limitation of tesla M10. In the datasheet the maximum is 64 users or vGPU instances.
How many xenapp sessions in vGPU or vDGA mode can host on one GPU board? Does the limitation apply for xenapp sessions as well?
How many virtual application ccu licenses do I need, 500?
Could following example a valid configuration for one esxi host with one tesla m10 board?
4 VMs with XenApp 7.13/Windows Server 2016 for ~120 sessions (hosted desktops)
vDGA Mode (with ESXi Standard License)
4x GPU Pass-through
I know that’s depends on user load and only a test drive is the best answer. However, I’d like to understand if this a hardware or a licensing limitation.
You’ll need a vApps license for each connected session if you’re only deploying the Apps and not full desktop. So if you have 500 CCU, then you’ll need 500 vApps licenses.
Those results don’t apply to your deployment anyway, so don’t worry about them.
As for how you’re going to deploy XenApp, use vGPU, not Passthrough. If you use Passthrough, you’ll need a vWS license for all users. So assign an 8A vGPU profile to each of the 4x XenApp servers. Give each XenApp server plenty of RAM and CPU and see how you get on. Depending on your CPU choice (speed / cores) and RAM allocation, you should be able to hit 30 CCU per XenApp host.
What’s your CPU model and generation? How much RAM is in the server?
If it were me, I’d opt for this CPU "E5-2697A v4" and spec each XenApp as follows to start, and then tailor RAM and CPU up or down accordingly:
CPU: 8 vCPU
RAM: 32GB
vGPU: 8A
The E5-2697A v4 has 16 physical cores (it’s a great CPU choice for many different workloads, if you have the option, then I’d go for this!). So when you’re assigning 8 vCPU to each XenApp VM, you won’t be over committing the physical CPUs. Remember, with ESXi, memory is hard allocated to each VM with a GPU when active. So no memory sharing between ESXi and the other VMs. You should be looking at 192GB RAM in the server. With all 4 XenApp VMs on, plus ESXi, you’ll have committed RAM of well over 128GB.
I’m a little bit confused about vApps license and full desktop. In the KB of nvidia I found the answer that you’re able to deploy full desktops with the vApp license.
[i]Q: If I use XenApp to publish full desktop is that licensed as applications (vApps) or as Virtual PC?
A: For all usage from an RDSH solutions such as XenApp the license required is vApp. As long as the profile associated to the VM is either A profile, or passthrough.
In the event that you choose to deploy Pro-Vis applications via RDSH / XenApp and you require Quadro functionality you would then require a vWS license and a Q profile or passthrough associated with the VM.[/i]
My favorite is to use vGPU. Before your very good recommendation, I planned with 14Cores Intel Xeon E5-2690v4 and 256GB RAM. So, I consider your recommendation at the design.
Sorry, my mistake. I misread the original post. As you say, yes, you’re able to deploy an RDSH desktop or Published App using the vApps license, up until you need a Pro-Vis App due to its feature requirements.
That CPU should be ok, but obviously has 2 less physical cores to make use of per socket, so just be aware of that when you’re trying to allocate CPU resource equally between the 4 VMs. Nice chunk of RAM, no problems there. I think you’ll be CPU limited before you run in to any other resource limits on those specs with that amount of users, obviously depending on what applications are being used.
I agree and follow your recommendation. 16 cores are the right size for these four XenApp VMs.
It exsists a PVS on bare metal with direct attached stroage for an existing xenapp 6.5 environment on hyper-v with quadro 4000. This will be a migration project to vsphere with tesla on the same pvs. Unfortunately, I don’t have the relevant hardware configuration like RAM and RAID Level yet. Do you see generally any constraints on the same pvs?
For me, having deployed many of each, MCS is by far the easier to use, especially with lower density user environments such as this. MCS is built into the product, requires no additional hardware, windows licensing or resilience requirements. PVS on the other hand, is the complete opposite.
Also, the days of requiring dedicated, physical PVS server are long behind us. These should be virtual if you intend to go down this route.
If your XenApp VMs are all going to be identical, then you don’t need a PVS environment to support 1 Master Image. It would be a waste of resource and add complexity that you simply don’t need.
Something that will make your life easier as well, when designing your XenApp environment, make sure you use AppDisks (free and built-in) or if you have the appropriate Citrix licensing - App Layering (UniDesk) so that you separate the applications from the Master Image.
I’m sure you’re following the Citrix Synergy event, make sure you use the recently released 7.14, not 7.13. The newer version has some nice feature enhancements, not to mention GPU monitoring built in which will obviously be very beneficial to you …