Standalone ESXi Build Help Needed.

B

bmeehan

n00b
Joined
Mar 4, 2012
Messages
8
Some advice please..

I'm currently running vmWare workstation on a Core 2 Duo (12 GB RAM) laptop with 256 GB SSD.

I'm running 4 Windows Server 2008 R2 VMs that are used to build images (additional VMs) for deployment. These four servers include:
SQL Server 2008​
Active Directory, DHCP, DNS, Etc​
Deployment Server -- Microsoft Deployment Toolkit, Windows Deployment Services​
Gateway -- Internet Access, otherwise other VMs are on internal Host only network​

The problem I'm having is that when I spin up a couple of image builds (new VMs), it just way too slow (CPU is pegged). I'm thinking that an ESXi build might solve this problem. I've run this configuration on our production Hosts (Dell R710's) using the internal 15K SAS drives and although it comsumes a lot of CPU, it flies in comparison to it running on my laptop. I really don't want to run this in production, but need more horsepower.

Here is my planned configuration:
1 - LIAN LI PC-V352B Black Aluminum MicroATX - want micro ATX for portablity​
1 - SUPERMICRO MBD-X9SCM-F-O LGA 1155 Intel C204 Micro ATX Intel Xeon E3 Server Motherboard​
2 - Kingston 8GB (2 x 4GB) 240-Pin DDR3 SDRAM ECC Unbuffered DDR3 1333 (PC3 10600) Server Memory Model KVR1333D3E9SK2/8G​
1 - SeaSonic M12II 520 Bronze 520W ATX12V v2.3 / EPS 12V v2.91​
1 - Intel Xeon E3-1270 Sandy Bridge 3.4GHz LGA 1155 80W Quad-Core Server Processor​
1 - Crucial M4 CT512M4SSD2BAA 2.5" 512GB SATA III MLC Internal Solid State Drive (SSD)​
1 - Verbatim TUFF-'N'-TINY 4GB Flash Drive (USB2.0 Portable) Model 96815​
1 - Western Digital Caviar Black WD2002FAEX 2TB 7200 RPM SATA 6.0Gb/s​

I know this build doesn't have all the storage isolation, HA, VMotion goodies, but I need a fast machine to plow through the build of deployment images. The image deployment process moves a ton of data, hense the SSD. Once this project is done, I'll add the machine to our SAN and use it to dev / test new Production VMs.

The problem I've having is deciding on the above motherboard or switching to one that will allow me to up the memory to 32 GB (SuperMicro MBD-X8SIL-F-O). From my research the 2 x 8GB Kingston memory will not be available until at least May..:mad: and is way overpriced.

Has anyone compared these two motherboards?

Thanks for your advice / help.
 
Just a FYI and this might make your decision easier. The SUPERMICRO MBD-X9SCM-F-O is a socket 1155 which is the current socket vs the SuperMicro MBD-X8SIL-F-O which is socket 1156. So that cpu that you have picked out will only run on the X9. Also both will go up to 32gb memory....

edit****
Plus I would save some money and get a E3-1230.
 
I went with the MBD-X8SIL-F-O in my home lab to save on memory costs. Two of them are up and running with 16gb each (unbuffered) and they perform quite well.

If I ever need to go to 32gb, the option is there when registered comes down in price.
 
You're making me really wish I didn't load up on 4x4gb udimms already, been a while since I've checked on prices.
 
Thanks for the quick replies.

mayhem87
Yes, I understand that the two boards have different sockets. One is a Lynnfeild, and the other is a Sandy Bridge.

shnelson
The benefit to me of the X9SCM-F board is that it has the C204 chipset that supports 2 6GB sata devices. Since the intended use of this host is to build images, the faster the hard disk, the better.

In case your interested, the build process goes like this:
  1. Create a blank VM (install OS later)
  2. Powerup the VM, PXE boot the machine
  3. Windows Deployment Server pushes down WinPE image (small Windows 7 image) into memory
  4. Deployment toolkit presents dialogs for selecting the OS, naming computer, selecting applications to install etc.
  5. Once Dialog boxes are completed
  6. Formats VMDK disk,
  7. drops Operating System Install on the disk,
  8. Injects drivers based on Make & Model
  9. Boots machine, completes Install of OS
  10. Installs selected Applications
  11. Installs Windows Updates
  12. Sysprep's the image
  13. Captures the image for later deployment.

This gives me the images that I need to deploy to the desktops at work. The deployment process is basically the same, but the majority of the applications get delivered along with the image. As you can see this is very intensive on the network and disk -- There is a vast amount of data being pushed around. The virtual switches help with the network, the disk access and CPU utilization seem to be the bottlenecks.

I would love to setup a couple with external storage for all the HA, Vmotion goodies. I envy your setup.

Netwerkz101
Yes the memory is what has been disappointing. Hopefully someday (soon) unregisters 8GB DIMMS will be available and more reasonably priced.

I agree the sweet spot is the E3 1230; it is the least expensive processor that supports hyper threading. I'm going to change my configuration to this model.

Thanks again for all your help
 
shnelson said:
You're making me really wish I didn't load up on 4x4gb udimms already, been a while since I've checked on prices.
Click to expand...

But you are in a best case scenario right now because your board can use 8GB registered dimms which are the cheap ones.

32GB (4x8GB) ecc registered = ~$270+
32GB (4x8GB) ecc unbufferred=~$600+
 
bmeehan said:
Thanks for the quick replies.

<snipped>
In case your interested, the build process goes like this:
  1. Create a blank VM (install OS later)
  2. Powerup the VM, PXE boot the machine
  3. Windows Deployment Server pushes down WinPE image (small Windows 7 image) into memory
  4. Deployment toolkit presents dialogs for selecting the OS, naming computer, selecting applications to install etc.
  5. Once Dialog boxes are completed
  6. Formats VMDK disk,
  7. drops Operating System Install on the disk,
  8. Injects drivers based on Make & Model
  9. Boots machine, completes Install of OS
  10. Installs selected Applications
  11. Installs Windows Updates
  12. Sysprep's the image
  13. Captures the image for later deployment.

This gives me the images that I need to deploy to the desktops at work. The deployment process is basically the same, but the majority of the applications get delivered along with the image. As you can see this is very intensive on the network and disk -- There is a vast amount of data being pushed around. The virtual switches help with the network, the disk access and CPU utilization seem to be the bottlenecks.
<snipped>
Click to expand...

That's pretty slick!! I wish I had the patience to build that ..... I guess I will get into that when I start playing with XenDesktop deployments.

You aren't interested in buying a more capable laptop, are you? I may be selling mine.
It wouldn't compare to an R710 with RAID, but it's much easier to carry around. ;)
 
It actually gets slicker when we do the actual deployments to the desktops. I work in a school district and over the summer we clean the computers (blow out the dust bunnies), and wipe and reload the os. We have to to approx 2000 machines. We prep a sql server database with the what we want on the machine and record this information by the machines MAC address.

When the cleaners finish they PXE boot the machine and the process starts automatically by looking up the computer MAC address in the database. It even puts a shortcut ini a folder that will remotely connect to the computer during the deployment process. It also reports progress on a console.

Oh and in case your wondering we multicast so the network doesn't get buried with traffic.

As for the laptop, No thanks. I'm satisified with my Dell Percision Moble Workstation M6400.
 
It actually get a little slicker when we do the actual deployment to the desktops. I work in a school district and over the summer we have to clean (disinfect and blow out the dust bunnies) and re-image over 2000 computers.

We prepopulate the SQL database for each computer based on the computers MAC address. When the cleaners finish they PXE boot the machine and the deployment automatically starts. It even puts a shortcut in a folder so that I can remotely connect to the computer during deployment. We also multicast the images so the network doesn't get buried.

As for the laptop, no thanks. I'm quite happy with my Dell Precision Moble workstation M6400. The R710 are what we use to host our production VMs.
 
With all of that considered, I would stick with the X9SCM and take advantage of the cheaper registered memory. My choice with the X8SIL was somewhat unique in the aspect that I have no need for local storage.

Though I have yet to see an MLC SSD fail as direct attached storage to ESXi, I would caution that with continuous writes like that it may be a good idea to pursue an SLC instead to avoid any surprises (failure) in the middle of a massive deployment/image build.


You mention there are 2000 machines that ultimately need to be deployed to, but how many unique images does that require you to create?
 
Shnelson:

The x9SCM only supports unregistered memory, and you just can't get 8 GB unregisted DIMMS. So I think I'm going to try to get it done with 16 GB (4 x 4GB). Registed DIMMS won't work on this motherboard.

As for the number of images, that's very flexible.

Using other technologies we had to build an image for each unique hardware configuration. Now that the the appropriate drivers are injected into the machine based on Make and Model before WIndows setup starts. We don't need images for each model. Everytime we get a new model of computer, I just have to build and test a driver library, and update the database to associate the make and model with this library.

The number of images needed is a function of:
  1. The OS being installed. We obviously have to have an image for each OS (XP, XP x64, Win 7, Win 7 x64)
  2. The applications that we want deployed along with the image.
Additional applications can be installed during the actual deployment. These usually include:
  1. Specialized drivers (again based on Make and Model -- information stored in the database by Make and Model)
  2. Applications that don't handle being syspreped.
  3. One Off applications that are only needed on a few machines.
The time consuming part is that all applications must be able to be installed silently (sometimes they have to be repackaged). Once the application installations are automated it's really just a function of updating the databases to define which applications are associated with each role. The deployment images are then built (this is what I need this ESXi host for -- Building images is very resource intensive (it just takes too long on my vmWare workstation).

To answer your question, we build about 10-15 deployment images, and do the remaining customization at deployment time.

Finally, one interesting part of this, is the captured images can be combined into a single WIM file which uses Single Instance Storage. This removes duplicate information from multiple images.

The machine I'm building is for creating these deployment images. I'll use the production Host for the actual deployment. So I'll stick with the MLC SSD (nightly backups of course).
 
My bad, it appears I got a little dyslexic last night!

I still think you'll be fine with your choice for board. I would recommend doubling your initial amount of memory to 16gb if you are doing a lot of simultaneous building. You'll burn up 8gb pretty quick with just a couple VMs and their overhead. Even if it is just temporarily the contention will cause performance degradation for the expansion & deployment process. If it's one build at a time then don't worry about it.

If you're interested in seeing a benchmark of the X8SIL in action, let me know which metric(s) or scenarios you'd like as I can throw an SSD in and give it hell. I have one box sitting idle with no VM's on it so it'd be a good candidate.
 
You must log in or register to reply here.
Back
Top