Results 1 to 16 of 16
  1. #1
    Join Date
    Feb 2003
    Posts
    44

    50 million page faults in 12 hours

    I have recently changed to a new Windows Server 2003 Standard Edition based server.

    When I first got the server, performance was really slow. I thought I figured it out when I noticed one of the 512mb memory sticks was PC2700 whereas the server spec was PC3200 for all 3 512mb sticks.

    However, even with it replaced with PC3200 memory, the server is still slow. It is quicker than before though - pages actually load within 10 seconds or so, opposed to over a minute.

    The stats for the server show 50 million page faults for a 12 hour period. I know page faults are part of computers, but does 50 million sound right?

    The server spec is 2.8ghz P4 HT with 1.5GB PC3200 DDR and has about 300-800 people browsing within 15 minute period.

    Is this large number of page faults normal? I never checked the amount of page faults on my old server and unfortunately it's too late now.

  2. #2
    Join Date
    Jan 2005
    Posts
    296
    I'd get the ram and hard drive checked out sounds like this could be a hardware issue. Also you said you just changed to this server, did the old server run ok and how comparable are the specs of the new one to the old one?

  3. #3
    Join Date
    Feb 2003
    Posts
    44
    Thanks, I'll do that. The old server specs were very similar - 3.2ghz P4 HT with 2GB PC2100 DDR. It ran fine with no problems at all. CPU usage was minimal, however it's maxing out on the new server.

  4. #4
    Join Date
    Aug 2004
    Location
    Zurich, Switzerland
    Posts
    770
    Using PC3200 vs PC2700 DDR should make no visible difference when rendering webpages, sounds like memory module or memory controller (on the mainboard) issues, the speedup was probably a result of getting rid of one of the faulty modules.

  5. #5
    Join Date
    Jul 2002
    Posts
    1,441
    2.8 1.5 GB RAM

    vs

    3.4 2 GB RAM is not very similiar, there is a margin of performance there. Also, what are the hard drives? SATA? IDE? SCSI?
    Synergy Blue LLC
    SonataWeb.net | SynergyBlue.com
    USA should so something about: http://www.brillig.com/debt_clock/

  6. #6
    Join Date
    Feb 2003
    Posts
    44
    Here's the server spec.

    Both drives are 7,200rpm IDE.
    Attached Files Attached Files

  7. #7
    Join Date
    Aug 2004
    Location
    Zurich, Switzerland
    Posts
    770
    Try removing the memory module in the socket in the middle for a day or two and see whether the number of page faults greatly decreases, it's not impossible since most mainboards aren't working well with all memory sockets filled. Best population is to use #1 and #3 if it's a 3 or 4 sockets board.

    BTW doesn't the chipset support dual channel memory? If yes, filling all three sockets you automatically disable that and reduce memory throughput by 50%.

    At any rate, neither the CPU clock nor the memory speed nor the memory size difference between the two listed configs should create any visible effects on webpage processing unless the server is heavily overloaded. If this is the only site running on the server, the upper limit of 1-1.5 million/month visitors shouldn't be a prob for this hardware, a server with much lower specs could handle it.

    Wait a moment, I just realize my estimations were for a Linux box, they might not apply for a Winblows machine. Anyway, it sounds more like faulty hardware than a software config prob.

  8. #8
    download and run a program called memtest86, usually a bootable ISO or floppy. This will tell you if there are any problems with the memory
    http://www.reloadedhosting.com
    Email me: [email protected]

  9. #9
    Join Date
    Feb 2003
    Posts
    44
    The memory has been tested and I have been told by the data center that it's "not faulty".

    The chipset supports dual channel operation, I have requested that they remove the 512mb DDR module from Channel-A DIMM-1 or replace it with a pair of 256 DDR modules in Channel-A DIMM-1 and Channel-B DIMM-1.

  10. #10
    I did a little searching and this might give you a little more explanation.

    http://computerperformance.co.uk/Hea...ory_Health.htm

    From reading that, you can draw the conclusion that page faults is not a memory error, but a report that your server is swapping to disk. This may indicate that you need more memory.

    What was your previous platform's OS?
    *AlphaOmegaHosting.Com* - Hosting since 1998
    Managed Dedicated Servers and VPS
    Hosted Exchange 2010 Email Service

  11. #11
    Join Date
    Aug 2004
    Location
    Zurich, Switzerland
    Posts
    770
    Second option's no good, that could cause further problems and system instability (different size memory modules).

  12. #12
    Join Date
    Mar 2005
    Location
    United States
    Posts
    308
    Switching from PC2700 to PC3200 shouldn't cause that dramatic change in computer perfomance based off the memories frequency...probably a bad module like someone above stated.

    The pc3200 should of automatically down-clocked itself to match the PC2700 (333mhz).

    Most page faults are due to memory mapping errors. Can you tell me if these faults your machine is logging are causing reboots?

    Try removing sticks of memory one by one to see if you can find the stick causing the issue. There may also be a problem with the OS assigning memory addresses.

  13. #13
    Join Date
    Feb 2003
    Posts
    44
    It turns out the memory was not tested - I checked the server logs. I informed the data center and I was just told it is scheduled for later today.

    dandanfirema: The previous OS was also Windows Server 2003 Standard Edition.

    RambOrc: Are you sure? If I understand Intel's web site correctly (http://support.intel.com/support/mot.../CS-012556.htm - scroll down to the diagram of the memory), the server should be able to handle 2 modules of PC3200 512mb and 2 modules of PC3200 256mb and still perform in dual-channel.

    Redcoat: There hasn't been any reboots since using Windows, but there was a regular reboot every few days when I first got the server and it was running CentOS. I thought it was a problem with the OS/install at the time. The server only had 1GB (2x512mb) of memory when it had CentOS loaded, but I'm not sure how the two modules were placed in the motherboard, or whether one of those were the PC2700 module that was removed.

  14. #14
    Join Date
    Aug 2004
    Location
    Zurich, Switzerland
    Posts
    770
    Sure it's still running in dual channel mode, that's not what I meant. I was talking about possible problems because pretty much all mainboards and chipsets since the beginnings were performing optimally only when not mixing memory modules of different make and size. Another thing that sadly also applies to the majority of mainboards that they work better if only half the memory sockets are filled (e.g. 2 out of 4). I don't say this necessarily applies to your specific mainboard in question, but chances are rather high that it does. I've seen too many otherwise great systems become instable and faulty once memory was upgraded from 2 to 3 or 4 modules. Of course I'm talking about the average mainstream mainboard and not about the high-end server ones, but I don't think there are too many data centers out there who would lease high-end servers for those 100-300 bucks/month when at IBM for example the low-end servers start at more than 5'000 with dual Xeon, 2 GB RAM and SCSI RAID 1.
    Last edited by Orc Webhosting; 05-24-2005 at 12:58 PM.

  15. #15
    Join Date
    Jan 2004
    Location
    North Yorkshire, UK
    Posts
    4,163
    Get all the memory swapped out, it does sound like a memory problem, but I'd assume if it was bad RAM the box should just fall over completely rather than being slow. A CPU problem is also a possibility.

    Is it recognising all of its RAM?
    How much RAM is shown as free during idle?
    Are there any specific processes shown as using lots of CPU?

    Dan

  16. #16
    Join Date
    Feb 2003
    Posts
    44
    After a week of late nights trying everything...

    ...it turned out to be the BIOS! (ev91510A.86A.0213)

    Apparently the Intel 915 chipset running the 0213 BIOS has a serious performance problem.

    Anyway, it's running flawlessly now:

    Average CPU usage: 34%
    Average Memory usage: 22%
    Average System Disk Queue: 0.038
    Average Database Disk Queue: 0.175

    Page faults are still high, but it doesn't seem to be affecting performance.

    Anyone running the 915 chipset, I strongly recommend you update your BIOS if you're running version 0213. You will see a HUGE performance difference.

    Thanks everyone for your help

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •