High-Def Video on VMware View 5.0 with BTL & 3D

While trying to find and identify technical benefits implemented in PCoIP protocol in VMware View 5 I have been taking note of interesting things about the behavior of the protocol that might be interesting for my readers. The average administrator probably won’t have a need to get deep into the technology, however if you are like me you will probably enjoy reading the results of the tests.

I have previously run a test comparing PCoIP and ICA on iPad. Read at

Comparing VMware View and XenDesktop on iPad.

For this test I used the High Definition sample video that comes with Windows 7 and produced results based on amount of video RAM assigned to a VM while changing the BTL (Built-to-Lossless) state. If you are not sure about BTL, I recommend reading 

What’s New in VMware View 5.0 (beyond Marketing).

The number of displays and resolution are the same across the tests below, however these values will only determine the amount of vRAM assigned to the VM. What I’m trying to say is that the vRAM is a number of MB resulting from the combination between numbers of displays and resolution. In VMware View 5 there is a new option: Enable 3D. If this option is enabled VMware View will not use displays and resolution to calculate vRAM, and instead it will just accept the amount of vRAM in MB assigned by the administrator.

 

 

In order to try to discard bandwidth variations across tests the HD video was executed three times for each scenario. The statistics were collected using the PCoIP WMI implementation.

This setup consists of a notebook connected over a TCP based VPN (Cisco AnyConnect) over my wireless hotspot at my home. The connection has 20Mbit downstream to the internet and the VPN gateway is located on the US West Coast, which is where I am based at.

The simulated configurations you see in the table below.

image

Note: To see a descent difference while using 3D, the vRAM was set to the limit of 128MB.

 

From the results above it’s possible to conclude:

Enabling 3D with 128MB vRAM helps to reduce kbit/sec peaks. Both samples with 3D enabled demonstrated considerable reduced peaks while the sample with BTL disabled demonstrated reduction of 38% when comparison to no 3D and BTL enabled.

On the average the sample with BTL disabled and no 3D was the winner with the least kbit/sec (4831.006). Interestingly, the second best was the sample with Lossless and 3D enabled.

 

My Comments:

The sample with BTL disabled and 3D enabled has the worst average bandwidth consumption while has the best reduced peak consumption. This make sense – while trying to reduce peaks the protocol will increase averages.

If users are in a RoBo (Remote Office/Branch Office) with properly sized bandwidth without burst capacity you will need reduced peaks, and the option with 3D and BTL disabled suits best this use case.

If users work remotely from unknown bandwidth size such as hotels and 3G/4G then the option with 3D disabled and BTL disabled suits bets due to reduced average bandwidth consumption.

 

My tests prove that the use of the new 3D feature in vSphere 5 and VMware View 5 will not only serve to take advantage of desktop graphics enhancements provided by AERO (such as peek, shake, and Flip 3D), DirectX9 and OpenGL 2.1, but also directly influence how PCoIP deliver content to end-points. The simple action of enabling 3D with 128MB reduced the average kbit/sec.

The combination of 3D and BTL is a powerful tool for administrators. As Uncle Ben would say – with power comes responsibility.

 

Below you will find results for each sample and will be able to see how PCoIP protocol  behaves differently each time the video is ran based on factors like bandwidth, latency, packet loss or jittering.

 

Lossless (disabled) – 3D (disabled)

image

 

Lossless (enabled) – 3D (enabled)

image

 

Lossless (enabled) – 3D (disabled)

image

 

Lossless (disabled) – 3D (enabled)

image

6 comments

Skip to comment form

    • Kyle on 10/23/2011 at 8:26 am

    Did view 5 drop the recommendation of not using TCP VPN connections for PCoIP?

  1. @Kyle
    Due to the PCoIP UDP nature it’s recommended to use UDP based VPN or the PCoIP Gateway via Security Server.

    Andre Leibovici

    • Kyle on 10/23/2011 at 5:28 pm

    @Andre Leibovici
    I understand that, I was just making a note in your article for testing… you stated you used a TCP VPN connection.

    • puscavnik on 11/26/2011 at 10:27 am

    Can I ask you what was the configuration of VM?
    1 CPU, 1gb ram?

  2. @puscavnik
    I don’t recall it, but I believe it was 1vCPU, 2 GM RAM with Windows 7 32b. That’s usually my standard configuration for baselining my tests.

    Hope that helps,
    Andre

    • puscavnik on 12/20/2011 at 5:42 am

    I did read your article VMware View 5.0 3D Reverse Engineered, but I still don’t understand why enabling 3D would reduce average bandwidth consumption. I tried it at home and it really does that but why? Better compression because it uses more CPU power and RAM on ESXi?

    What is very interesting, if you enable “Configure the maximum PCoIP session bandwidth” the average bandwidth consumption will go higher than with out this limitation, but it reduces packet loss.

Leave a Reply