Performance on VDI (Windows XP Virtual Machines)

[timjim88]

We have 2 ESX servers running VMWare Infrastructure 3. There are 32 GB a piece in each of these servers. We are also running fiber between our SAN and ESX servers (All VMs are on the SAN only). We run 66 Windows XP VMs in this environment with Wyse V10L thin clients. The thin clients use RDP to connect to our VMs. This environment is in a K-12 corporation.

Over the summer I rebuilt 2 of our thin client labs (total of 40 VMs). When I say "rebuild" I am meaning that I took an existing VM (VDI-XP18 for this example) and brought it off the domain. I then cloned it and began updating it (the clone). I went ahead and put VDI-XP18 back on the domain and let it run again. At that time I was working on the clone (renaming it of course) and installing SP3 for WinXP. Once this was done I cloned the VM to another one. The reason for this is to create my second image for another lab. Once both of my templates (at the time VMs) were built up I converted them with VMWare back to templates. I then deployed each template to the prospective labs. Everything was working fine in the summer (Keep in mind that when I say "working" I mean that when I booted up each VM I could get into them (RDP and on VIC) and things worked fine. Now that students are back everything is SLUGGISH. I look at the processes within the VM and on the ESX hosts themselves. Everything appears normal. I would say there is about 50% utilization.

Does anyone have any suggestions or ideas on why things may be sluggish all of a sudden? Both labs have Service Pack 3 now. One lab now has Office 2007, whereas before it was 2003. The other lab has SynchronEyes 7.0 (SmartTech Technologies). I did power down ALL VMs at one point. When booting up one lab at a time things were still sluggish. The 26 VMs that were never touched this summer work perfectly fine, indicating that something has gone wrong when I made the changes to the VMs.

Any help is GREATLY appreciated! By the way, I did also ensure that the VMs are divided up on the VNICs. 50% is on VNIC0 and 50% on VNIC1. This goes for both ESX hosts. Thanks!

-t.j.

[azn2kew]

You can safely run View 3.x with 64 VMs per host between 512-1024MB ram and doesn't have any issue at all and assuming you have the right hardware/network infrastructure to support that. What type of 'SAN are you using and how you're architect/carve it? Do you have enough network bandwidth and storage memory cache? Seems like you have small VDI lab environment so you shouldn't have any problems with performance unless that XP18 template is having issue after SP3 upgrade. Have you try test to deploy new XP client and recompose and rebalance from there to see if you still have same performance. What type of slowness you're seeing....boot time, interactive slowness etc...to tweak performance, use GPO to disable unwanted services, screen saver etc...

If you found this information useful, please consider awarding points for "Correct" or "Helpful". Thanks!!!

Regards,

Stefan Nguyen

VMware vExpert 2009

iGeek Systems Inc.

VMware, Citrix, Microsoft Consultant

[Spencer14]

I would venture a guess that your bottle neck issue is i/o related on your storage. Give me a break down of what is currently your storage configuration (array model/make, disk group raid level, how many disks are in the group and speed of the disks).

[timjim88]

I'm not certain on the configuration on the SAN. A contractor installed this about a year ago. I have built a VM with this template. This VM runs Deep Freeze Enterprise Server. It hasn't had any issues, BUT it does manage roughly 150 to 200 physical client machines. I don't know if you're familiar with Deep Freeze, but the clients to send a small amount of packets every three minutes back to the server. This was running fine before summer, but during the summer I did point more clients back to the server. Is it possible that this network traffic is causing issues? I recently began pointing physical machines back to a newly built physical server, hoping to eliminate some of these issues.

The "slow" performance is primarily logging in (maybe 10 to 20 minutes per VM). We run the VMs 24/7, but I have configured Deep Freeze to reboot them over the weekends; they take a while to boot up when 67 VMs boot up simultaneously. As I mentioned above, I don't know the exact config on the SAN, but we have two ESX servers. They are HP ProLiant DL385 G5s running with dual quad-core AMD Opteron Procesoors (2352). Each one has 32 GB memory.

[VMmatty]

I think it's time to start looking at the performance statistics while you're seeing slowness in the XP virtual machines. This will hopefully help you narrow down where the bottleneck is occuring.

For something like this, I would go to the console of one of the ESX hosts and run esxtop in batch mode. This will allow it to capture data and record it to a CSV file which you can then read in Windows Performance Monitor. The command is the following:

esxtop -a -s 5 -b >output.csv

-a = All Counters

-s = Time interval for monitoring. In this example, gather data once every 5 seconds

-b = Batch mode with output file

Make sure you do this on a partition where you have enough space to store a file that will probably end up being several megabytes depending on how long you let it run. Once you have the data you can start looking at what is peaking during the period of poor performance and go from there.

Also, is there anything else different about the new XP VMs? Is there a scheduled antivirus scan? Scheduled defrags, etc?

[jpccastle]

Hello,

I can tell you I had a bad time getting thinclients to work with synchroeyes, the Smarttech support said they hadn't "certified" it. I would try uninstalling it and see if your performance improves.

John

[timjim88]

A while back I contacted SmartTech. They actually had me delete a Gina.dll file from the registry. I don't have my notes handy so I don't recall the exact location, but once that file is deleted (in Safe Mode was the only way I could do it) it ran fine. The clients had to be pointed to the hostname, not the Teacher ID. They never informed me that it wasn't "supported," but they did explain that was the only way to run SynchronEyes on VMs/Thin environment.

[timjim88]

The contractor who performed the initial install came and checked things out, and helped me resolve the issue. You mentioned GPOs to disable un-needed services; this turned out to play a factor. My image was not a good image to deploy. Here are a few things that DRASTICALLY increased performance. I did not defrag my images before storing the templates. I recently learned that this applies both to physical and virtual environments. I also turned off a handful of services and features before creating the template (i.e. wireless configuration wizard, system restore, page file size = 0, etc....). The engineer explained to me that even though something like wireless zero config wiz may not require resources, it does require a bit when you multiply it by X amount of VMs. Once I rebuilt the new slimmed-down templates and redeployed things are running VERY smoothly.

Tweaking my initial template was the answer.

[DFN]

Would it be possible for you to detail the exact things you tweaked?

[timjim88]

Here is the documentation I saved for myself to remember down the road:

Build VM from scratch:

Make sure you use LSI Logic and not BUS logic, as the LSI will run more efficiently on your I/O performance.

Services Disabled:

Themes

Wireless zero configuration

Event log

Indexing service

Disabled Startup Programs (typing in msconfig at the Run option in Start menu):

Windows Card Space

Other non-essential programs (may vary slightly depending on software…i.e. If you have Quicktime you may want to disable qttask, if you have iTunes you may want to disabled ituneshelper, etc……)

Power Settings:

• Right-Click on desktop > Properties > Screen Saver (tab) > Power…

o Turn off monitor: Never

o Turn off hard disks: After 20 mins

o System standby: Never

Run Disk Defragmenting

Run Disk Cleanup Utility

Disabled paging file:

• Right-click My Computer > Properties > Advanced (tab) > Settings (Performance area)

o Under the Visual Effects tab, select “Adjust for best performance”

o Under the Advanced Tab select Change

 Then select “No paging file” to disable the paging file altogether.

Notes: When disabling the paging file you may eventually find Virtual Memory Low errors popping up. It will all depend on the memory allocated to each VM.