Pauge faults question - Aix

This is a discussion on Pauge faults question - Aix ; We've got a shiny new AIX 5.3 box running websphere, and it's suffering somewhat. One interesting bit is if I look at Nmon (v11 or v10) I'm getting 20000 or so page faults per second. Nmon thinks this is a ...

+ Reply to Thread
Results 1 to 5 of 5

Thread: Pauge faults question

  1. Pauge faults question

    We've got a shiny new AIX 5.3 box running websphere, and it's suffering
    somewhat. One interesting bit is if I look at Nmon (v11 or v10) I'm getting
    20000 or so page faults per second. Nmon thinks this is a problem - is it?

    If it is a problem, any hints on what to do about it?

    There's enough RAM - we're not using more than we have yet.

    cheers,
    clive


  2. Re: Pauge faults question

    Clive George wrote:
    > We've got a shiny new AIX 5.3 box running websphere, and it's suffering
    > somewhat. One interesting bit is if I look at Nmon (v11 or v10) I'm
    > getting 20000 or so page faults per second. Nmon thinks this is a
    > problem - is it?
    >
    > If it is a problem, any hints on what to do about it?
    >
    > There's enough RAM - we're not using more than we have yet.


    If you're page-faulting, then you _don't_ have enough RAM, at
    least for the workload. If you don't want any page faults, that is.

    You may want to cut down the size of the file cache (which can
    be up to 80% of real memory) to leave more for running processes.
    I don't recall the options to adjust this, but I believe you will
    use vmo.

  3. Re: Pauge faults question

    Gary R. Hook wrote:
    > Clive George wrote:
    > > We've got a shiny new AIX 5.3 box running websphere, and it's suffering
    > > somewhat. One interesting bit is if I look at Nmon (v11 or v10) I'm
    > > getting 20000 or so page faults per second. Nmon thinks this is a
    > > problem - is it?
    > >
    > > If it is a problem, any hints on what to do about it?
    > >
    > > There's enough RAM - we're not using more than we have yet.

    >
    > If you're page-faulting, then you _don't_ have enough RAM, at
    > least for the workload. If you don't want any page faults, that is.


    Hi Gary,
    i my understanding the - Page Fault - rate in nmon displayes the
    amount of page - including files - not found in memory.

    So a - page fault - without any - to Paging Space - means more or
    less that a file could not be found in memory and thus will be loaded
    from disk or a newly created file will be written to disk.
    Depending on the application this behaviour might be as aspected - like
    with TSM and maybe with dynamic http data creation as well.

    So to the OP.
    A performance snapshot of the system is not really a good baseline for
    performance tuning.
    So check the way websphere is creating/reading/writing file, get the
    tuning information for websphere and verify these setting, check what
    else runs on the system and do a long term measurement of your server
    load.

    >
    > You may want to cut down the size of the file cache (which can
    > be up to 80% of real memory) to leave more for running processes.
    > I don't recall the options to adjust this, but I believe you will
    > use vmo.


    If the page fault stays high and no paging to paging space occures i
    would increase the file cache or like you said by more ram.

    cu
    Hajo


  4. Re: Pauge faults question

    "Hajo Ehlers" wrote in message
    news:1161683656.033363.257770@k70g2000cwa.googlegr oups.com...



    It appears to be the ldap server (tivoli directory server) generating a lot
    of these - what do people think about moving this to another machine? Would
    one normally run an ldap and websphere on the same box?

    cheers,
    clive


  5. Re: Pauge faults question

    Clive George wrote:
    > "Hajo Ehlers" wrote in message
    > news:1161683656.033363.257770@k70g2000cwa.googlegr oups.com...
    >
    >
    >
    > It appears to be the ldap server (tivoli directory server) generating a
    > lot of these - what do people think about moving this to another
    > machine? Would one normally run an ldap and websphere on the same box?
    >
    > cheers,
    > clive

    I am running into this on my WebSphere box. In a nutshell, you have two
    different applications that need to be tuned almost in an opposite
    manner, running on the same box. In addition to the LDAP server on mine
    I also have the database that my WebSphere app uses residing on the same
    box. I inherited this mess, and if I have to make it different in the
    future, I will separate these out. I am also seeing the high rate of
    system calls and no paging utilization. I interpret this as everything
    is pinning itself in memory. DB2 does it, and I think the java app in
    WebSphere is doing it.

    May one's stupidity serve as a warning shot to others! ;-)

+ Reply to Thread