Topic
5 replies Latest Post - ‏2013-04-05T21:44:55Z by nagger
SystemAdmin
SystemAdmin
2404 Posts
ACCEPTED ANSWER

Pinned topic About PROC(Runqueue,syscall,swap-in) and LPAR(sys%) in NMON

‏2013-03-15T03:57:41Z |
Hi all. I am analyzing one of the clients' application running on AIX. I found in some consecutive days the peak CPU shown in NMON got lower and compared to other days the syscall/swap-in is higher while the runqueue is higher in the peak CPU usage period(syscall/swap-in/runqueue are all in PROC sheet).
And in other days the higher part than the lower days of the peak CPU is "sys%" but I cannot find evidence in TOP sheet.
So could you pls tell me what syscall/swap-in higher while runqueue lower means and what is the relationship of syscall/swap-in and runqueue?
And could you pls advice where I can find the process so I can find what the sys% cpu usage is in peak CPU period in NMON?
Thank you first!
And here goes the snap shot of NMON data
Updated on 2013-04-05T21:44:55Z at 2013-04-05T21:44:55Z by nagger
  • MarkTaylor
    MarkTaylor
    70 Posts
    ACCEPTED ANSWER

    Re: About PROC(Runqueue,syscall,swap-in) and LPAR(sys%) in NMON

    ‏2013-03-23T17:53:58Z  in response to SystemAdmin
    what nmon command flags, count and sample rate are you using ?
    does this happen frequently ? on certain days only ? or just random ?
    grab 1 second data, with at least -t and limit top proc data collection with something like -I 10, then upload the nmon file so we can take a look (make sure you remove any sensitive data first though if required etc)

    Rgds
    Mark Taylor
    • SystemAdmin
      SystemAdmin
      2404 Posts
      ACCEPTED ANSWER

      Re: About PROC(Runqueue,syscall,swap-in) and LPAR(sys%) in NMON

      ‏2013-03-27T13:25:35Z  in response to MarkTaylor
      Thank you Mark! Sorry for late response. Actually it is a batch server to handle file, doing compressing etc. I am working on the file uploading difference between two days these days.
      The attached is the raw nmon file. Thanks for you great help!
  • nagger
    nagger
    1620 Posts
    ACCEPTED ANSWER

    Re: About PROC(Runqueue,syscall,swap-in) and LPAR(sys%) in NMON

    ‏2013-03-25T20:48:36Z  in response to SystemAdmin
    CPU Sys% is the time in the AIX Kernel either due to System calls or device drivers.
    The Run Queue is the number of processes that are able to run at the rime.
    Swapping is due to virtual memory.

    These are only loosely related.

    Our comments tend to show that the workload was quite different on the two days.
    Like running different batch jobs or a different mix of workloads but not much else.

    These are side effects rather than a problem or bottleneck.

    cheers, Nigel Griffiths
    • SystemAdmin
      SystemAdmin
      2404 Posts
      ACCEPTED ANSWER

      Re: About PROC(Runqueue,syscall,swap-in) and LPAR(sys%) in NMON

      ‏2013-03-27T13:32:48Z  in response to nagger
      Thanks Nagger.
      The server is a batch server to handle file uploading and compress them. I check the file uploaded those two days. The frequecy and the size has not difference to cause that large gap on peak CPU usage. I want to find some evidence and hints from the NMON file.
      • nagger
        nagger
        1620 Posts
        ACCEPTED ANSWER

        Re: About PROC(Runqueue,syscall,swap-in) and LPAR(sys%) in NMON

        ‏2013-04-05T21:44:55Z  in response to SystemAdmin
        Hi,
        I have looked at the nmon files.
        The 121204 one is doing loads more read disk I/O and has 25 GB free memory - where the other file has 12 GB.

        Was the machine restarted before the 121204 and the memory flushed of all cached disk pages.

        The processes also have less memory 48% - the other data is 62%
        numperm not 10% - other sheet 22%
        Run queue peaks to 16 - other sheet is 48

        So some differences but these show perhaps a different initial start position in what is memory.
        It does not explain why it is different.

        Hope this helps, thanks, Nigel Griffiths