locked
Disk performance drops when using Microsoft Deployment Tools RRS feed

  • Question

  • I posted this to the Win8 Forum to

    Hi

    So using the same iso if I mount a virtual disk through the drac (dell 7910) and install win 8.1 i get a disk read of about 250MB/s using winsat no matter how many times i reboot it stays around 250MB/s however if i use the same iso and convert it into a wim file using MDT then use the default client task sequence to deploy the disk read time is around 150MB/s. If I remove all the post deployment options in the task sequence it will start at 250MB/s until you reboot at which point its back down to 150MB/s. What gives?

    Friday, May 1, 2015 8:41 PM

Answers

All replies

  • The difference between 250MB a second and 150MB for actual deployment is not really a big deal. You will go from 5 minutes to 7.5 or so.  That being said if your concern is deployment time then this may be of interest to you:
    http://blogs.technet.com/b/deploymentguys/archive/2015/03/27/reducing-windows-deployment-time-using-power-management.aspx?pi36647=1

    Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.

    • Proposed as answer by Ty Glander Monday, May 4, 2015 6:37 PM
    Monday, May 4, 2015 6:37 PM
  • Thanks Ty, I hadn't seen that post. Have you tried it? What kind of real world differences were seen?

    If this post is helpful please vote it as Helpful or click Mark for answer.

    Monday, May 4, 2015 7:31 PM
  • Yes it is a pretty good difference (physical hardware because VM is really as per the host instead)

    Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.

    Monday, May 4, 2015 8:26 PM
  • Its not deployment time its the response time on the client after receiving the image. 250MB/s vs 150MB/s on someones workstation is a huge deal if they are doing things that are drive intensive. The drivers are identical in version and date. We have not made a diff of the regkeys yet. I should also point out that i can reproduce this behavior on 100% of the dell 7910s I have tried (about 30 in all) 
    That behavior is if you use an iso you'll get 250 mb/s read time on the target machine after install but if you use the default client task sequence you'll get 150 mb/s on the target machine after install.

    • Edited by phyvtab Monday, May 4, 2015 9:14 PM
    Monday, May 4, 2015 9:06 PM
  • There is nothing I am aware of that would effect the disk performance post install.

    What version of MDT?  What ADK (or maybe WAIK if you are using old stuff)?

    When you say converting the ISO to WIM with MDT what are you meaning?

    Can you give detailed steps? There is quite a bit of ambiguity in what you are doing with MDT.

    What OS are you using?

    Is it a custom image or is it publicly available media?

    Are you doing a standard client task sequence?

    Make\Model?


    Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.

    Monday, May 4, 2015 9:21 PM
  • I answered most of that above but I will try to be more clear.

    MDT 2013, standard client task sequence no changes made. The ISO is win 8.1 off of MSDN nothing custom. This is a totally new MDT/ADK/WDS server which had been working "fine" for months until we happened to install once manually and saw the performance boost. 

    I used the import OS option on MDT with full set of source files. Also if I pick a WIM image from WDS and use it through MDT vs just WDS only I get the same behavior. Whenever we add MDT into the mix it causes the target computers drive speed to slow down. 

    This 8.1 is deployed using a standard client task sequence no custom scripts or changes. The OS deploys successfully but we only get 150mb/s disk performance on the target computer after this deployment is made. Using WDS with the same WIM file or using the original MSDN ISO we get 250mb/s 

    I used to have MDT fully automated installs off of pxe but i have pretty much turned off everything I added at this point and I am back a default install while seeing the same behavior. 

    These are dell 7910s, the driver pack is loaded from the dell site here http://www.dell.com/support/home/us/en/19/Drivers/DriversDetails?driverId=PMGW6

    Monday, May 4, 2015 10:20 PM
  • It might be interesting to look and see what drivers are installed on the MDT installed OS versus the other installed OSes. powershell: http://www.verboon.info/2014/04/how-to-export-third-party-driver-packages-using-powershell/

    When you are injecting drivers are you allowing MDT to detect drivers or are you doing something similar to?: http://deploymentresearch.com/Research/Post/325/MDT-2013-Lite-Touch-Driver-Management


    Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.

    • Marked as answer by phyvtab Tuesday, May 5, 2015 9:27 PM
    Monday, May 4, 2015 11:42 PM
  • Right now im letting MDT detect the drivers. I agree it feels like a driver issue but even when I took two machines and made one with the ISO and one with MDT and made a diff of the drivers, then manually installed drivers to match i ended up with the MDT one still being 100mb/s slower. Theres a few other things ive thought of that I havent had time to investigate yet like maybe its not really defaulting to NEWCOMPUTER, some regkey setting, and the auto generated unattended file. It would appear that if I interrupt the MDT OS install right before it finishes and hits the next phase step it will infact then be 250mb/s and stay 250mb/s no matter how many reboots. But it leaves in a computer in a dirty state. If you allow to continue as far as Next Phase we find the bad behavior.  Since this is before the post install I'm going to see if I notice anything odd in the unattend file today. 
    Tuesday, May 5, 2015 4:26 PM
  • Bitlocker maybe?

    Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.

    Tuesday, May 5, 2015 5:40 PM
  • Not Bitlocker, got the same issue with 2012r2 on these machines and the bitlocker feature is not installed when complete. Hoping to try against some other hardware to. 
    Tuesday, May 5, 2015 6:50 PM
  • That being said if your concern is deployment time then this may be of interest to you:
    http://blogs.technet.com/b/deploymentguys/archive/2015/03/27/reducing-windows-deployment-time-using-power-management.aspx?pi36647=1
    Ty, I just wanted to add that I used the information from the link (with modifications to make it work in MDT w/o SCCM) and I immediately have noticed 20-50 minutes knocked off real world deployment times so far. Thanks!!

    If this post is helpful please vote it as Helpful or click Mark for answer.

    Tuesday, May 5, 2015 8:57 PM
  • If you could post bdd.log and setupapi.dev.log to something like Onedrive.com from your MDT install.

    Also the setupapi.dev.log from one of your setup.exe installs will be interesting.

    As far as I am aware there shouldn't be anything that would cause a perf difference with MDT.


    Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.

    Tuesday, May 5, 2015 9:26 PM
  • So it looks like having all the drivers in a single bucket was bad. I made some selection profiles per your link and the initial results look good. 
    • Proposed as answer by Dan_Vega Tuesday, May 5, 2015 9:34 PM
    Tuesday, May 5, 2015 9:28 PM
  • I've never liked having tons of drivers mixed together as is slows down deployments (injected unneeded drivers) and can sometimes lead to hardware getting a driver that wasn't the best match. I use Johan's Total control method which Ty referenced MDT 2013 Lite Touch Driver Management. That method hasn't failed me.

    If this post is helpful please vote it as Helpful or click Mark for answer.

    Tuesday, May 5, 2015 9:33 PM
  • I've never liked having tons of drivers mixed together as is slows down deployments (injected unneeded drivers) and can sometimes lead to hardware getting a driver that wasn't the best match. I use Johan's Total control method which Ty referenced MDT 2013 Lite Touch Driver Management. That method hasn't failed me.

    If this post is helpful please vote it as Helpful or click Mark for answer.


    And frankly pnpenum doesn't really handle hardware with a parent child relationship (where the child device doesn't show up until after the parent driver is installed).

    Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.

    Tuesday, May 5, 2015 10:04 PM