NVIDIA
Sizing Question - XenAPP 7.14 / Tesla M10
Hello, I'm new to this topic with VMWare, Citrix and GPU Virtualisation... We are currently at the new acquisition of servers. One Tesla M10 should also be installed in these units. The servers are equipped with 2x E5-2690v4 and 384 GB of RAM. ESXi 6.0 is to run on them, and XenAPP 7.14 within the Windows 2016 VMs. On each server will be run 8-10 VMs with 4 or 6 vCPUs & 32GB RAM. Is this a usable configuration for about 160-200 Office users (Office 2016 / IE11 ...) per server? Which Profile and License is most suitable for 2 monitors a 1920x1200 per user? Regards Thomas
Hello,
I'm new to this topic with VMWare, Citrix and GPU Virtualisation...
We are currently at the new acquisition of servers. One Tesla M10 should also be installed in these units. The servers are equipped with 2x E5-2690v4 and 384 GB of RAM. ESXi 6.0 is to run on them, and XenAPP 7.14 within the Windows 2016 VMs. On each server will be run 8-10 VMs with 4 or 6 vCPUs & 32GB RAM.
Is this a usable configuration for about 160-200 Office users (Office 2016 / IE11 ...) per server?
Which Profile and License is most suitable for 2 monitors a 1920x1200 per user?

Regards
Thomas

#1
Posted 06/01/2017 12:58 PM   
Hi Thomas Firstly, make sure you use ESXi 6.5, not 6.0. There are GPU related performance issues with 6.0 that are resolved in 6.5. You don't mention how many physical servers you have? Regards Ben
Hi Thomas

Firstly, make sure you use ESXi 6.5, not 6.0. There are GPU related performance issues with 6.0 that are resolved in 6.5.

You don't mention how many physical servers you have?

Regards

Ben

#2
Posted 06/01/2017 02:21 PM   
Hello Ben, which are performance problems in vmware 6.0? Are there KB articles of VMware? Regarde Thomas
Hello Ben,

which are performance problems in vmware 6.0?
Are there KB articles of VMware?


Regarde

Thomas

#3
Posted 06/02/2017 10:48 AM   
Hi I'm not aware of any KB articles. This is personal experience, but also confirmed in conversations with VMware's vGPU team. I'm not going to list every issue I've experienced with 6.0, however the big one for me that is repeatable across different customer platforms, was a huge Power On / Shutdown time for vGPU enabled VMs with large memory. With 6.5 this is [u]dramatically![/u] reduced. Plus, while you're waiting for the VM to start or shutdown, you can do absolutely nothing with the host while it deals with the RAM. This applies for "Power On" and "Shutdown" (where memory is allocated to the VM, and then released back to ESXi). Restarts are fine as the RAM has already been allocated to the VM and doesn't get released until the VM is Shutdown. Regards Ben
Hi

I'm not aware of any KB articles. This is personal experience, but also confirmed in conversations with VMware's vGPU team.

I'm not going to list every issue I've experienced with 6.0, however the big one for me that is repeatable across different customer platforms, was a huge Power On / Shutdown time for vGPU enabled VMs with large memory. With 6.5 this is dramatically! reduced. Plus, while you're waiting for the VM to start or shutdown, you can do absolutely nothing with the host while it deals with the RAM. This applies for "Power On" and "Shutdown" (where memory is allocated to the VM, and then released back to ESXi). Restarts are fine as the RAM has already been allocated to the VM and doesn't get released until the VM is Shutdown.

Regards

Ben

#4
Posted 06/02/2017 01:49 PM   
Hi Ben, They can still say something about the planned configuration and License profil Regarde Thomas
Hi Ben,

They can still say something about the planned configuration and License profil


Regarde

Thomas

#5
Posted 06/06/2017 08:36 AM   
Hi Thomas Sorry, I don't understand that last post? Can you elaborate please ... Regards
Hi Thomas

Sorry, I don't understand that last post? Can you elaborate please ...

Regards

#6
Posted 06/06/2017 12:31 PM   
Hello, sorry for the late reply. I was prevented the last time. My question was whether you can say something about my dimensioning question and the necessary licensing. The problem with VMWare I will clarify. Furthermore, I still have a question. How does XenaAPP assign the GPU resources to the user. I assign the VM a profile with 1GB frame buffer. Is this framebuffer then used for all users or is each logged-in user a share of 1GB of the entire physical framebuffer of the GPU? Will I get only 32 users (with an M10 and 1GB framebuffer) on a physical server with accelerated graphics? Regards Thomas PS: Sorry for my bad english - I use Google :-)
Hello, sorry for the late reply. I was prevented the last time.

My question was whether you can say something about my dimensioning question and the necessary licensing. The problem with VMWare I will clarify.

Furthermore, I still have a question. How does XenaAPP assign the GPU resources to the user. I assign the VM a profile with 1GB frame buffer. Is this framebuffer then used for all users or is each logged-in user a share of 1GB of the entire physical framebuffer of the GPU? Will I get only 32 users (with an M10 and 1GB framebuffer) on a physical server with accelerated graphics?

Regards Thomas

PS: Sorry for my bad english - I use Google :-)

#7
Posted 06/22/2017 11:04 AM   
Hi Thomas With a XenApp VM, all resources are shared between the users on that VM. This is the same as CPU, RAM, Storage, Network and GPU. If you allocate a 1GB Profile, all users will share that 1GB Profile. For XenApp VMs, typically you'd assign an 8GB Profile. The reason you'd do this is two-fold. Firstly, you will have multiple users using the same resource, so there needs to be enough of it to go around. Secondly, when you allocate an 8GB Profile, you effectively "lock-out" the physical GPU (the bit that actually does the processing) for the users on that XenApp VM, so that no other VMs can access it. For example, if you allocated a 4GB Profile from the same GPU to 2 XenApp VMs, all users on both XenApp VMs would have access to the same single physical GPU, and depending on the user density and applications being run, you may have degraded performance. And if (again, for example) you have 25 users on each XenApp VM, then that's 50 users that have access to the same GPU! (which is potentially too many). As for the 32 users, yes, with a 1GB Profile a single M10 will support 32 users, but you'll be using XenDesktop to do this, not XenApp. Although "technically", you could use XenApp, but there wouldn't be any point in doing that :-) Regards Ben
Hi Thomas

With a XenApp VM, all resources are shared between the users on that VM. This is the same as CPU, RAM, Storage, Network and GPU. If you allocate a 1GB Profile, all users will share that 1GB Profile.

For XenApp VMs, typically you'd assign an 8GB Profile. The reason you'd do this is two-fold. Firstly, you will have multiple users using the same resource, so there needs to be enough of it to go around. Secondly, when you allocate an 8GB Profile, you effectively "lock-out" the physical GPU (the bit that actually does the processing) for the users on that XenApp VM, so that no other VMs can access it. For example, if you allocated a 4GB Profile from the same GPU to 2 XenApp VMs, all users on both XenApp VMs would have access to the same single physical GPU, and depending on the user density and applications being run, you may have degraded performance. And if (again, for example) you have 25 users on each XenApp VM, then that's 50 users that have access to the same GPU! (which is potentially too many).

As for the 32 users, yes, with a 1GB Profile a single M10 will support 32 users, but you'll be using XenDesktop to do this, not XenApp. Although "technically", you could use XenApp, but there wouldn't be any point in doing that :-)

Regards

Ben

#8
Posted 06/22/2017 12:13 PM   
Scroll To Top

Add Reply