locked
Server 2012 - advice on high Page Faults RRS feed

  • Question

  • Hi All

    I have a 2012 Server running Hyper-V for a few VM's. It's a good spec Dell Poweredge with 192GB RAM.

    Typically it runs around 130GB RAM available. Pagefile is managed by system, currently allocated 25,600 MB.

    I am currently testing Veeam One monitoring solution and one of the things it's highlighted for this server is that the Page Faults/sec is high and above the warning threshold. I'm seeing alerts with values of 3,606.0 and 2,1560.0 Pages/sec.

    I have since run perfmon on the server and tried to make sense of the data I'm seeing.

    I just paused Perfmon and have the following values:-

    1. Page Faults/sec - Last 285.452, Avg: 1,066.649, Min: 1.002, Max: 15,149.914 

    2. Page Input/sec - Last 0.00, Avg: 0.654, Min: 0.000, Max: 33.051 (I believe this is the number of reads from pagefile.sys)

    3. Transition Faults/sec - Last 139.221, Avg: 528.399, Min: 0.000, Max: 8,423.310 (I believe this is the number of soft faults)

    Strangely (to me anyway), I would have expected the number of Page Input/sec + Transition Faults/sec should = Page Faults/sec?

    Do these figures look correct and normal to you all?

    If not, where do I start with diagnosing the issue?

    Thanks in advance for help and assistance given :)
    • Edited by PMThomas Thursday, April 21, 2016 1:45 PM
    Thursday, April 21, 2016 1:45 PM

Answers

  • I don't think your numbers are out of line at all.  Over what period of time are you viewing these stats?

    "Page Input/sec - Last 0.00, Avg: 0.654" means that you are going to disk less than once a second, and remember that when the pager goes to disk, it is often reading lookahead to minimize the number of hard reads.  So this value can often be inflated over what is really needed.

    I tend to manually configure my page file to 1-4 GB because Hyper-V needs very little page file.  There are spikes if I log onto the system and do something interactively, but when it is running with no interactive logins, it rarely does any hard paging.

    https://technet.microsoft.com/en-us/library/cc781216(v=ws.10).aspx - Monitoring Paging.  Written for 2003, but the memory management variables have not really changed.


    . : | : . : | : . tim


    • Edited by Tim CerlingMVP Friday, April 22, 2016 2:13 PM
    • Proposed as answer by Hello_2018 Tuesday, April 26, 2016 9:10 AM
    • Marked as answer by Hello_2018 Thursday, April 28, 2016 3:06 PM
    Friday, April 22, 2016 2:12 PM

All replies

  • Hi PMThomas,

    >>Strangely (to me anyway), I would have expected the number of Page Input/sec + Transition Faults/sec should = Page Faults/sec?

    The page input plus transition faults often does not equals page faults.

    It is hard to simply plus these parameters with a simple formula.

    Besides, I suppose the following blog might help us to better understand page file related issue:

    http://social.technet.microsoft.com/wiki/contents/articles/13383.best-practices-for-page-file-and-minimum-drive-size-for-os-partition-on-windows-servers.aspx

    https://technet.microsoft.com/en-us/magazine/ff382717.aspx

    Additional resource:

    Evaluating Memory and Cache Usage

    https://technet.microsoft.com/en-us/library/cc958290.aspx

    Best regards,


    Andy_Pan

    Friday, April 22, 2016 7:10 AM
  • Thanks for the response Andy.

    The 2nd article is more focussed on hard page faults, when it's having to go back to the page file on disk.

    My issues seems to be soft page faults, where it's looking somewhere else in memory. Are my figures unusually high? They look it to me, certainly compare to other systems we have here.

    Friday, April 22, 2016 1:39 PM
  • I don't think your numbers are out of line at all.  Over what period of time are you viewing these stats?

    "Page Input/sec - Last 0.00, Avg: 0.654" means that you are going to disk less than once a second, and remember that when the pager goes to disk, it is often reading lookahead to minimize the number of hard reads.  So this value can often be inflated over what is really needed.

    I tend to manually configure my page file to 1-4 GB because Hyper-V needs very little page file.  There are spikes if I log onto the system and do something interactively, but when it is running with no interactive logins, it rarely does any hard paging.

    https://technet.microsoft.com/en-us/library/cc781216(v=ws.10).aspx - Monitoring Paging.  Written for 2003, but the memory management variables have not really changed.


    . : | : . : | : . tim


    • Edited by Tim CerlingMVP Friday, April 22, 2016 2:13 PM
    • Proposed as answer by Hello_2018 Tuesday, April 26, 2016 9:10 AM
    • Marked as answer by Hello_2018 Thursday, April 28, 2016 3:06 PM
    Friday, April 22, 2016 2:12 PM
  • Hi Tim

    This was just a random point in time.

    Looking at the server now, Transition faults/sec is Avg: 1,014 and max: 22,633.

    To me the Max looks very high. But is it compared to everybody else out there?

    As it happens, I have just changed the pagefile on the Host server to 1GB and will restart it over the weekend. Typically RAM is 65% Free.

    Friday, April 22, 2016 2:32 PM
  • Hi PMThomas,

    >>As it happens, I have just changed the pagefile on the Host server to 1GB and will restart it over the weekend. Typically RAM is 65% Free.

    After you changed the host server's memory, did this performance improved?

    If you have any updates, welcome to share here.

    Best regards,


    Andy_Pan

    Tuesday, April 26, 2016 9:14 AM