How to increase graphical performance of view desktops

[vxaxv17]

We are starting deployment of Win7 enterprise virtual desktops in our organization using 23" LG all in 1 zero clients (pcoip).  They are running at 1920 x 1080 native resolution.  We have 20 VM's running right now on 3 esxi servers (192GB ram, dual 6 core xeon processors). The server hardware is barely being taxed at all. storage is on ibm v7000 FC san with SSD drives.  This is a very robust setup which can run many more than the 20 VMs we are running. VMs have 2gb ram and 2vcpu.

The problem we are having is graphical performance is pretty bad.  Windows are slow when resizing/dragging, video playback is choppy, scrolling in internet explorer stutters and we have one particular application that lags tremendously when switching menus.

We've noticed that if we set the zero clients to a lower (non-native) resolution of 1024x768, then the performance increases greatly and is comparable to a physical workstation.

Why is this?  Does vmware view (5.1) really not work well with these larger monitors and higher resolutions?  Would graphics offload cards help at all?  Im struggling to find where exactly the problem is and how it is best resolved.  I've looked at the teradici apex cards but from what I have read, it doesnt sound like they give better performance so much as take the burden off the server CPU which is definitely not a problem at this point.  Server CPUs are barely above 5%.

[LMUISHuntsville]

vxaxv17,

You are correct about the Apex offload card.  It does not sound like it would benefit you much in your situation.

With that said take a look at this link and apply the reg setting it describes http://blogs.vmware.com/performance/2012/10/turbo-charge-view-video-performance.html.  This should help and hopefully remedy your issue.

Regards,

Nathan

[DavidPatrick]

We evaluated two APEX 2800 cards for 6 months and found they are not worth the investment.

They will provide virtual machine consolidation on the hosts as per their documentation.

We have been looking closely at the nVidia VGX cards. http://www.nvidia.com/object/grid-vgx-software.html

Grid Boards http://www.nvidia.com/object/grid-boards.html

http://arstechnica.com/information-technology/2012/10/nvidias-vgx-cards-bring-big-graphics-performan...

Cheers 

[LMUISHuntsville]

Yes, while using GRID or Quadro 4000 - 6000 GPUs will increase 3D application performance via vSGA it will not help much at all with playing videos.  Currently as I understand it 2D HD video playback is not hardware accelerated by vSGA.  vSGA is only targeting 3D applications for office productivity, visualization, CAD, and some games based on either DirectX 9c or up to OpenGL 2.1. 

Regards,

Nathan 

[bjohn]

Have you tried increasing video memory?

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=103100...

[ericblc2]

Are your PCOIP clients Tera2 or Tera1 ?

It feels to me like you have something that is not right. Here is my experience.

With a similar setup as yours (kick ass esx ressources + view 51 + zero client) I was able to get very good display performance with tera2 client except for full screen video.  Scrolling internet explorer or playing with menus was never an issue. Mind you my VMs were all on SSD in the host and were using much more ressources (in the lab). Tera 1 will be more sluggish than tera 2. That's for my lab.

Now for our pilot (in production):

Right now in our pilot we have 75 users (tera2 zero clients) spread over two ESX  hosts. The VM are Windows 7, 32 bit, 2vcpu and 1.5 gig ram. We are using linkclones on local SDD (replica and checkpoint vmdk on local SSD). The users (zero clients), the user data (file servers) and the view/esx infrastructure are in 3 different location in the city (gig+ speed wan).  The pcoip bandwidth is capped at 5 Mbs per VM. We get zero complaints about general display performance (excluding video). With a volontary 5 Mbs cap on pcoip we obviously limit video capabilities but if you watch youtube on a small window is still ok. Full screen is not usable.

We did start the pilot without local SSD in the hosts (backorder) and ran the VM on crappy shared storage SAN and performance was much worst (including sluggish logon time and windows menu navigation) but it was all fixed when we switched to SSD.

The problems you are describing are more severe than what we are experiencing which leads me to believe you have one component in your setup that is impacting your performance.  Not sure which one it is.

Hope that helps set your expectations.

Eric

[CyberTron123]

well, I can say for sure that  a vSGA is accelerating video, windows (if using aero in Win7) , and so on. I can now run fullscreen video (from youtube for example) inside of IE (but that is because IE and flash uses hardware gpu if availible. 3d games works perfect and so does autocad.

If I turn off hardware gpu in IE it is sluggish in full screen, also if I change to software it is also sluggish.

With that said, I am using a quadro 4000 card right now and will probably buy a GRID K1 soon. Overall performance is ok, not desktop performance yet, but it is close! (at least on lan)

but they still have a bit to go until I am completly happy althogh I wouldn't change VDI for anything!

[vxaxv17]

All of our devices are Tera1, is that the problem?  I'd love for it to be that simple.

Thanks

[kslish]

CyberTron123, can you elaborate on what kind of hardware you are using for your host server setup?  We are actively looking at swtiching our current environment (HP blades) to something that can support GPU cards.

vxaxv17, Tera 2's should theoretically improve performance because of the better chipset and the fact that they can use client side caching to decrease the bandwidth requirements.  However, if it's a network issue (bad QOS rules limiting PCoIP UDP traffic for example) they won't change much.

I've had excellent results with 1280x768 resolution 19" monitors and zero clients, and "acceptable" performance using 1680 x 1050 resolution 22" monitors and zero clients.  I've stayed away from 1920 x 1080 resolution monitors due to the drop off in performance since the higher the resolution, the higher the bandwidth needed to update all those pixels and we unfortunatly aren't Gigabit ethernet at all of our locations yet.

[CyberTron123]

Hi

we are serving around 80 concurrent users with 3 Fujitsu RX350 with 2 cpu (6-cores) and 160gb ram

these are connected to  Netapp NFS shares

I am currently testing a Nvidia quadro 4000 card with good 3d and overall performance, however we are probably buying nvidia GRID K1 for production.

with the quadro around 10 people are able to use 3d acceleration,

We can play Open Arena at 1280x720 with no latency issues, or watch fullscreen youtube trailers and use autocad. So it is good, and it does improve graphics. However, for general Windows 7 one has to use AERO to get the hardware accelerated windows, and they do have a bit left there i belive, don't get me wrong, there is an improvement, but it is not perfect yet!

[admin]

For the use case you are describing an APEX is not going to do much and you also likely do not need vSGA ( Hardware Assistance ).

With higher resolutions like your describing the biggest thing is likely the VRAM. If you have the pool setting set for exactly that display set up you might not have enough VRAM to drive that screen and get better performance. A Tera2 will help give you even better performance but your Tera 1 should be fine.

If you do not have a View version that lets you change the VRAM with the slider. Just increase the num displays to the max number ( 4 ) and highest resolution. Power off and on the VMs and then see if that helps.

WP

[vxaxv17]

We have ordered 2 tera2 demo units to compare and see if there is any noticeable difference in performance between the tera1 and tera2.  Will report back afterwards.

[DavidPatrick]

We have been using the DELL WYSE P25 units (Tera2) for about two months now and have found no real desktop performance enhancements over the Leadtek VP200P units (Tera1).

The P25 units run a lot cooler and the USB 2 devices performance is much better.

For real enhancement in desktop performance I would recommend you look at installing a GPU in your host e.g NVidia Quadro 4000 we are so far really impressed with the desktop feel and performance.

Maybe have a look at the NVidia Grid cards http://www.nvidia.com/object/grid-boards.html

DELL WYSE P25 http://www.wyse.com/products/cloud-clients/zero-clients/P25

Cheers.

[vxaxv17]

Thank you for the post.  Can you elaborate on how you define "desktop performance"?  Do you mean 2d functions like window dragging/resizing, window scrolling, keyboard/mouse input, etc or do you mean things like 3d performance and or video playing?

I get conflicting info about installing GPU cards in the esx hosts.  Some people say you get no display performance improvement, just greater VM density.  Others state that you do get a performance gain.  Cards are expensive to purchase just to try and see what happens.  And we have 3 esx hosts so we would need 3 cards.  Also, how does it work are there a certain # of VMs you can run per card?

Thanks.

[DavidPatrick]

The overall performance is much better but we are in the early stages of the evaluation process.

The feedback we are getting so far from our students and teachers is the following:

Adobe CS6 Photoshop 3D openGL is now working with 1-2s lag.

Accessing media "online content" through a web browser is more responsive.

Playing of film media at 720p or 1080p is now possible.

General mouse movements within the desktop is much more responsive.

We currently have 18 hosts and I am looking at maybe providing a GPU for about 4 of those hosts which deliver desktop services to our Web Design and Advertising classes.

I currently have two VDI pools which are identical "Web Design" running Adobe CS6 applications and one host has the GPU installed and the other does not.

The GPU Pool has 15 virtual machines which are set for hardware rendering and are using the GPU (Quadro 4000).

We did go down the path of evaluating the two APEX 2800 cards which as stated would provide consolidation "greater VM density" in the host but provided no user enhancements.

I would recommend you look at a NVidia Quadro 4000 which is one of the support cards and with minimal investment.

Also look here for reference for configuration settings http://communities.vmware.com/message/2224936#2224936

I hope this helps.

Cheers

David.

[ericblc2]

DavidPatrick wrote:

The feedback we are getting so far from our students and teachers is the following:

Adobe CS6 Photoshop 3D openGL is now working with 1-2s lag.

Accessing media "online content" through a web browser is more responsive.

Playing of film media at 720p or 1080p is now possible.

General mouse movements within the desktop is much more responsive.

Do you peg all these improvements to the addition of the Quadro card ?

Ever since I have started tinkering with vmware view and Teradici clients I have tried to find ways to make simple videos (ex: Youtube) watchable full screen. 

How does the video performance on a teradici client compare to a PC when connecting to a VM that is "helped" by a Quadro card in ESX?

Ebaying for a Quadro

[DavidPatrick]

Do you peg all these improvements to the addition of the Quadro card ?

Short answer is yes, but we are still in very early stages of evaluating this type of technology.

The cards we are looking at purchasing for production would be the NVidia GRID K1 & K2 models but that will be a considerable investment per server and we would need to determine the return on that investment.

So far we are getting very similiar results from both the Tera1, Tera2 and PC clients.

Cheers

[vxaxv17]

Can someone please quickly explain how its determined how many VMs you can run on a single quadro card?  Is it just a function of how much of the cards ram is allocated to each VM?

Thanks.

[admin]

It really depends on several things.

The amount of vRAM on the card is a general guideline and one hard limit. Keep in mind only 1/2  the vRAM comes from the GPU. For example, a VM configured for 512MB will only use 256MB from the GPU

Number of displays and resolution size plays a role. The more you use, the larger they are, the more it takes to drive them

Ultimately, it really comes down to workload. Is it basic AERO desktops or reasonably heavy 3D with high framerates is it DirectX based or OpenGL? Each one of those factors plays into the performance or utilization.

The GPU also plays a role. We have generally seen better performance from the Quadro line. However GRID has more VRAM and aligns well for basic desktop 3D usage, like Aero.

A few other points based on other comments that have come across.

- You will not get better video performance from vSGA. Today video is not accelerated by the GPU. For video intensive environments we recommend 2D based desktops with 2 vCPUs this is the highest performance, most cost effective approach today when using host rendered video.

- Adding GPU acceleration is not about getting a higher consolidation ratio. We do not see higher consolidation not for basic 2D desktops or even Aero and lighter weight 3D. You can get a better user experience depending on your applications. It really is more about expanding the possible use cases especially applications that can benefit from GPU acceleration such as CAD/CAM designers etc..

- The performance you get is directly assosited with the client capabilties today as well. PCoIP hardware zero clients provide the best overall performance for 3D based workloads again depending on your apps the Tera2 will provide better performance over the Tera1 through a higher sustained framerate. This might not be completely noticeable by all your users, but it is measurable.

WP

[vxaxv17]

Thanks for the info.

Our main focus is reducing the perceived lag that users are experiencing moving from a physical PC to a zero client w/VM.  We are not using any 3D applications.  We are talking about basic windows functions like moving the mouse, dragging windows, resizing, etc.  These are basic operations that while not slow, are noticably lagged compared to a phyiscal PC.  We have tera1 clients right now and have ordered a few tera2 clients to see if it helps any but it really sounds like we should be trying out a quadro card as well.