GRID M6 bandwidth requirements/observations
Has anyone had any experience measuring PCoIP session bandwidth using GRID? I'm in the process of running up a testing platform using GRID M6, M630 blades, Horizon 7 and Windows 7 virtual desktops. Point to point connectivity will be via PCoIP. Typical bandwidth expectations put forward by Teradici/VMware, particularly looking at Teradici's PCoIP bandwidth calculator, refer to an 'Extreme User' as the highest potential consumer of bandwidth. During my early stages of bandwidth recording, I see this 70-100Mbps area being reached using a variety of non-CAD apps, such as video playback up to and including 1080p, web browsing using Chrome/IE and Office application usage of Excel, Word, etc. I wouldn't refer to these applications as indicative of an extreme PCoIP user and am keen to know if others have seen similar measurements using GRID and PCoIP My testing setup is: 3 x 1920x1200 panels Win7 VM with 4 vCPU (E5-2660), 40GB storage, 16GB vRAM Horizon 7.0.2 (agent and direct connect plugin) ESXi 6.0.0 build-4192238 GRID M6-1B profile GRID 369.71 guest driver GRID 367.64 ESXi host driver Teradici zero client running 4.8 firmware
Has anyone had any experience measuring PCoIP session bandwidth using GRID?

I'm in the process of running up a testing platform using GRID M6, M630 blades, Horizon 7 and Windows 7 virtual desktops. Point to point connectivity will be via PCoIP. Typical bandwidth expectations put forward by Teradici/VMware, particularly looking at Teradici's PCoIP bandwidth calculator, refer to an 'Extreme User' as the highest potential consumer of bandwidth. During my early stages of bandwidth recording, I see this 70-100Mbps area being reached using a variety of non-CAD apps, such as video playback up to and including 1080p, web browsing using Chrome/IE and Office application usage of Excel, Word, etc. I wouldn't refer to these applications as indicative of an extreme PCoIP user and am keen to know if others have seen similar measurements using GRID and PCoIP


My testing setup is:

3 x 1920x1200 panels
Win7 VM with 4 vCPU (E5-2660), 40GB storage, 16GB vRAM
Horizon 7.0.2 (agent and direct connect plugin)
ESXi 6.0.0 build-4192238
GRID M6-1B profile
GRID 369.71 guest driver
GRID 367.64 ESXi host driver
Teradici zero client running 4.8 firmware

#1
Posted 03/20/2017 02:28 PM   
Noticed some further points - the View Agent (pcoip_server_win32.exe) service operates between 20-25% CPU when using a GRID enabled VM. Use the same VM without vGPU present - using the software option in the VM properties page (3 x 1920x1200 display, no 3D support, using the recommended amount of RAM for my displays and resolution) I see a far lower CPU usage for pcoip_server_win32 - around 5-10%. GRID must chew a lot of CPU resource as part of the software encode/decode
Noticed some further points - the View Agent (pcoip_server_win32.exe) service operates between 20-25% CPU when using a GRID enabled VM. Use the same VM without vGPU present - using the software option in the VM properties page (3 x 1920x1200 display, no 3D support, using the recommended amount of RAM for my displays and resolution) I see a far lower CPU usage for pcoip_server_win32 - around 5-10%. GRID must chew a lot of CPU resource as part of the software encode/decode

#2
Posted 03/21/2017 05:31 PM   
Hi Nealus, not really sure what you are trying to tell us. Everything you "observed" so far is clear and true. The more screen capture (higher FPS) you have the more encoding you need. Therefore we recommend NVENC. NVENC only works with Blast protocol. Concerning your bandwith requirements you won't get a proper response from anyone. This really depends on your workload, amount of monitors/resolution, required FPS, Horizon policies (image Quality) and so on. With PCoIP it is quite easy to fully leverage a 100Mbps for a single user. Why not testing with Blast? Regards Simon
Hi Nealus,

not really sure what you are trying to tell us. Everything you "observed" so far is clear and true. The more screen capture (higher FPS) you have the more encoding you need. Therefore we recommend NVENC. NVENC only works with Blast protocol. Concerning your bandwith requirements you won't get a proper response from anyone. This really depends on your workload, amount of monitors/resolution, required FPS, Horizon policies (image Quality) and so on. With PCoIP it is quite easy to fully leverage a 100Mbps for a single user. Why not testing with Blast?

Regards

Simon

#3
Posted 03/22/2017 07:04 AM   
I'm not testing with Blast because the endpoints I use only allow PCoIP. Believe me, I would love to see the benefits of H264/NVENC on the VMs CPU usage, but can't. The question is more along the lines of 'does anyone see an uplift in either CPU usage and/or PCoIP bandwidth when using a GRID enabled VM compared to the same VM running standard non-3D software graphics?' For me, the answer is yes to CPU usage, but no real noticeable increase in bandwidth with the same workload
I'm not testing with Blast because the endpoints I use only allow PCoIP. Believe me, I would love to see the benefits of H264/NVENC on the VMs CPU usage, but can't.

The question is more along the lines of 'does anyone see an uplift in either CPU usage and/or PCoIP bandwidth when using a GRID enabled VM compared to the same VM running standard non-3D software graphics?'

For me, the answer is yes to CPU usage, but no real noticeable increase in bandwidth with the same workload

#4
Posted 03/22/2017 08:46 AM   
Hi Nealus, thanks for clarification. To be honest I've not done such comparison yet but as I already wrote the increase in encoding is very clear, I would also expect an increase in bandwidth usage maybe not really recognizable with PCoIP as it always tries to use the available bandwidth and therefore the observations might become different when you test over WAN connection. Regards Simon
Hi Nealus,

thanks for clarification. To be honest I've not done such comparison yet but as I already wrote the increase in encoding is very clear, I would also expect an increase in bandwidth usage maybe not really recognizable with PCoIP as it always tries to use the available bandwidth and therefore the observations might become different when you test over WAN connection.

Regards

Simon

#5
Posted 03/23/2017 06:32 AM   
Scroll To Top