Horizon View 6.2 WAN performace over MPLS (3D Applications)

[ITVisionIT]

Hello,

Hoping the community here can assist and provide some suggestions or recommendations to improve my end user experceing using View and 3D Applications.  Below are the details of my environment.

View Host:

Cisco C240-M3S
ESXi 6.0.0-3620759
GRID K1 Card - Host Driver 361.45.09

VM Config:
Win 7
12GB RAM
2 vCPU
K160Q Profile

Application utilizing inside the guest OS of the VM is Autodesk Revit and Plant3D.

User Access:

User is accessing the VM using the 4.0.1 Horizon View Client over a 6MB MPLS connection.  Ping times from the uses location to where the VM is hosted is 15-17MS.

Endpoint Site:
5min Average traffic In: 275.2 kb/s
5min Average traffic Out: 174.1 kb/s.

Datacenter where the VM is hosted has a 50MB MPLS connection.

5min Average traffic In: 7675.9 kb/s
5min Average traffic Out: 29.6 Mb/s

Group Policy PCoIP Session Variables:

Set the Image Cache Size in Megabytes: 500MB
Set the Minimum Image Quality value (default value: 50): 50
Set the Maximum Initial Image Quality value (default value: 90): 100
Set the Maximum Frame Rate value (default value: 30): 100
Use image settings from zero client if available: disabled

Set PCoIP session bandwidth floor in kilobits per second to: 128
Enable/disable audio in the PCoIP session: Enabled
Enable/disable microphone noise and DC offset filter in PCoIP session: Enabled
Turn off Build-to-Lossless feature: Enabled

User experiences choppyness or delays when attempting to rotate the model in the 3D view.  I understand that tweeking these settings is an art.  I am not seeing any latency on the backend storage where this VM resides, which we are running a hybrid box with Fusion IO cards for QoS on the storage side.

Looking forward to suggestions from the community on helping improve my users experience utilizing 3D applications via VMware View.  I have also been looking at upgrading to View 7 to see if the Blast Protocol would provide better performance over the traditional PCoIP protocol.

Thank You

[TheLoneVM]

Everything looks good on the surface.  I suspect the issue may be that your workload is too much for the K1 to handle?  The K1 is typically recommended for "task worker" type environments, and the K2 or M60 is recommended for designers with high end 3D applications.  Do you get the same lag when using a less complex scene/model/environment using the same applications?  (IE: just create a simple box in Revit and rotate).  Upgrading to a combination of NVidia GRID K2 and View 7.0 with hardware h264 Blast would likely show major improvements.

[KISCharlie]

What does CPU utilization look like on the VM when thislaggyperformance is happening?

One thing that might be worth looking into is the use of a Teradici APEX card for PCoIP process offloading.   We have found that this can make a decent improvement in performance, especially in more demanding workloads.  The PCoIP process can eat up quite a bit of CPU resource when there are a lot of screen changes going on.

We've taken a VM running a 720p video in Windows Media player on a 1920x1080 resolution monitor and watched a dual core VM run at 80-95% CPU utilization with nothing else running.

That same VM running with an APEX card runs at 20-30% CPU utilization.

[RandyDGroves]

You may be overpowering your K1 with a 100fps sample rate. I would dial it back to 30 or even 24. Also, the image quality setting of 100 is quite bandwidth intensive requiring more data to be transmitted per frame which can lower the frame rate. I would suggest no higher than 90 and most people find 80 works fine (and even that is higher quality than Blast Extreme on most CAD content). Remember, the PCoIP image will build to perceptually lossless once it stops moving while Blast Extreme does not. The zero clients are limited to 250MB cache, so setting it to 500MB is just consuming memory in the VM. One final thing to check is whether the zero clients are reporting any packet loss during the session.

[ITVisionIT]

Thank you for the response.  We have used the APEX cards in the past.  I may look at re-implementing them again and see if they help.  I greatly appreciate the feedback.

I did make some adjustments to the VMs.  Bumped them from 2vCPUs to 4 and upped the GPU profile from 1GB to 2GB.  A few of the users have said they have noticed an improvement.

Thank You

[ITVisionIT]

Hi Randy,

Thanks for the info.  I think I will play around with these options a bit more as you suggested and see what comes of it.  We also utilize Riverbeds in our environment so I may looking see what optimization capabilities they can do for the PCoIP traffic.  In addition, look at applying some QoS settings for this traffic as well.  Currently our VoIP traffic is at the top of the list but I don't think we have any QoS setting for the View traffic.

Thanks for the response.

[ITVisionIT]

Thank you for the response.

We do have some view host that do have a K2 card installed.  You make a great point on the K1 vs the K2.  I will further troubleshoot performance and see what VM is using what host GPU card and see if I can get the users to try the performance of both.  Bumping up the VMs vCPU count from 2 to 4 and changing the GPU profile from 1GB to 2GB has help a little.

Thank you again for the feedback.

[RyanGoldstein]

I have 1 user on pass through VGPU on K2, he insists of having 16 vCPU.  I don't like it but it makes him happy and he says it works better...