NVIDIA
Grid / Horizon on the WAN
We're seriously considering moving forward with a Grid / Horizon solution where I work. Does anyone have experience with using this solution across the WAN for heavy CAD (Autodesk Inventor) use? We currently tested a vDGA box across a 10mb EPL and the results were acceptable but not great. LAN works awesome, fluid model pan/zoom/tilt. WAN has noticeable stutter. I've followed horizon best practices for optimizing the VDI environment. What kind of bandwidth do I need to throw at this thing to get models rotating smooth? I read 1mb for CAD use but it doesn't seem to cut it. Thanks!
We're seriously considering moving forward with a Grid / Horizon solution where I work. Does anyone have experience with using this solution across the WAN for heavy CAD (Autodesk Inventor) use? We currently tested a vDGA box across a 10mb EPL and the results were acceptable but not great. LAN works awesome, fluid model pan/zoom/tilt. WAN has noticeable stutter. I've followed horizon best practices for optimizing the VDI environment. What kind of bandwidth do I need to throw at this thing to get models rotating smooth? I read 1mb for CAD use but it doesn't seem to cut it.

Thanks!

#1
Posted 09/03/2015 02:50 PM   
Sorry, we're not a VMware shop, but I've been running some tests over the WAN to test out precisely this sort of thing with a (by today's standards) 15/1.5 Mb/Sec cable modem and get really good results with a Citrix solution, in particular incorporating their Framehawk technology. Latency was averaging typically around just 80 msec. This was running Citrix Receiver on an old laptop running Windows 7 connected to a XenApp 7.6 FP1 instance under Windows 2012 R2 and a GRID K2 back-end in passthrough mode. AutoDesk products like Revit, AutoCad, 3DS Max, etc. all ran really well. From the user perspective, I would say I was completely satisfied with the performance. -=Tobias
Sorry, we're not a VMware shop, but I've been running some tests over the WAN to test out precisely this sort of thing with a (by today's standards) 15/1.5 Mb/Sec cable modem and get really good results with a Citrix solution, in particular incorporating their Framehawk technology. Latency was averaging typically around just 80 msec. This was running Citrix Receiver on an old laptop running Windows 7 connected to a XenApp 7.6 FP1 instance under Windows 2012 R2 and a GRID K2 back-end in passthrough mode. AutoDesk products like Revit, AutoCad, 3DS Max, etc. all ran really well. From the user perspective, I would say I was completely satisfied with the performance.
-=Tobias

-=Tobias

#2
Posted 09/03/2015 06:30 PM   
1MB/s is an optimistic average, and not realistic when you're moving a model. A quick and very unscientific test provides the following: In Horizon VIEW, using a fullscreen CAD application on a 1920x1200 display rotating a model to locate a specific part or point on the model, the network spikes to 15 - 18Mb/s. I tried with 3 different applications and 3 different models and as you'd expect the results are the same for all. Also, same results when running wireframe or shaded. The thing to bear in mind though, is that these are burst values, and that when you have multiple users on the wire, it's unlikely they'll all be doing the same activity at the same time. When the model's not moving, it's using 0Mb/s so it you're taking a measurement, or checking materials, or some other motionless activity, then your usage drops. So, if you can give each user the chance to burst to 15-18Mb/s and measure utilisation across a link with multiple users working normally that would give you a better value. I'd suggest that realistically, for multiple users you could run closer to 5-8Mb/s as long as the capability to burts to 20Mb/s is present. Citrix HDX is more efficient across the wire, for the same test as above it was using under 3Mb/s, so if bandwidth is a limiting factor it may be worth taking a look at that, and as you can see from Tobias' results it could require even less.
1MB/s is an optimistic average, and not realistic when you're moving a model.

A quick and very unscientific test provides the following:

In Horizon VIEW, using a fullscreen CAD application on a 1920x1200 display rotating a model to locate a specific part or point on the model, the network spikes to 15 - 18Mb/s.

I tried with 3 different applications and 3 different models and as you'd expect the results are the same for all.

Also, same results when running wireframe or shaded.

The thing to bear in mind though, is that these are burst values, and that when you have multiple users on the wire, it's unlikely they'll all be doing the same activity at the same time. When the model's not moving, it's using 0Mb/s so it you're taking a measurement, or checking materials, or some other motionless activity, then your usage drops.

So, if you can give each user the chance to burst to 15-18Mb/s and measure utilisation across a link with multiple users working normally that would give you a better value.

I'd suggest that realistically, for multiple users you could run closer to 5-8Mb/s as long as the capability to burts to 20Mb/s is present.

Citrix HDX is more efficient across the wire, for the same test as above it was using under 3Mb/s, so if bandwidth is a limiting factor it may be worth taking a look at that, and as you can see from Tobias' results it could require even less.

Jason Southern, Regional Lead for ProVis Sales - EMEA: NVIDIA Ltd.

#3
Posted 09/07/2015 06:14 PM   
Scroll To Top

Add Reply