• 1 reply
  • Latest Post - ‏2013-03-25T22:51:37Z by nagger
3 Posts

Pinned topic nmon TOP and waiting processes

‏2013-02-27T14:00:23Z |
Hello all,

I've been using nmon & the analyser to investigate slow job turnaround times for jobs executed on a client system (a dedicated 9133-55A).

I've noticed high instances of IOWAIT over a period that corresponds quite nicely with two busy disks. So far, so good.
Now I want to nail down the specific processes in wait.

The TOP sheet in the nmon analyser output file shows the "big hitters" in terms of CPU%. (We are using the default threshold of 0.1% for identifying a "big" process)

In all cases, the (%user CPU + %sys CPU) = %CPU, suggesting that none of these processes is waiting.
My question is ... are waiting processes discarded/ignored by nmon ?
Thanks and regards,
Updated on 2013-03-25T22:51:37Z at 2013-03-25T22:51:37Z by nagger
  • nagger
    1738 Posts

    Re: nmon TOP and waiting processes

    You have funny ideas about Wait stats.

    Processes in "wait" are not using the CPU and are not on the run queue - they are asleep waiting on an interrupt queue (see ps -el WCHAN column).

    CPU Wait time is actually CPU idle time but with the waiting for disks I/O to return flag set.
    So no CPU is actually being used - actually it has to be doing something and that is running the wait process or yeilding a shared CPU to another LPAR and really not using the CPU.

    This Wait time is calculated against the CPUs during clock interrupts and not individual processes - because they are not doing anything.

    Processes not using CPU time in User mode (the application code) or System mode (running kernel system calls for the process) will be ignored, if they use less than the CPU threshold.

    Cheers, Nigel Griffiths