2 Replies Latest reply on Sep 13, 2012 12:57 PM by 960622

    Qacct output, help understanding values


      I would like some explanation on the output of the qacct command that will parse the sge accounting file. I'm using SGE version 6.2u5.
      When for example I ask for a statistical overview of my user, I get the following output: (put the || for overview)
      OWNER       WALLCLOCK         UTIME         STIME           CPU             MEMORY                 IO                IOW
      <username>        335639      1529.544      2875.904      4438.677    17179889002.853              3.586              0.000
      If I understand correctly, wallclock is the time in seconds spent for all the jobs. This is the time of the cpu spend in user time/sec (utime) and system time/sec (stime) counted together. So if I'm not wrong there already I get that much.
      But for the other values I can't seem to find an overview of what they mean. CPU, is that the percentage cpu, the amount of cycles, seconds? Same with MEMORY? Are those bytes, megabytes? And same for IO (read? write? per second?)

      Also, why is it, it doesn't seem to count IO to nfs mounts nor IOW, which I presume is io-wait? Because those 2 values would be very valuable for me. If this is not possible with qacct, is there a way to monitor nfs io per job using sge?

      I hope my questions are clear, but if not, do not hesitate to ask for more information!

      Kind regards,
        • 1. Re: Qacct output, help understanding values
          Satishbabu Gunukula
          did you see below links


          Why you should attend Oracle OpenWorld?
          You want to participate in Oracle Beta program or Exam?
          • 2. Re: Qacct output, help understanding values
            Thank you for the links. However they still leave me puzzled about the MEMORY and IO values. I don't get what is meant by "Gbytes cpu second". So the value of MEMORY divided by CPU gives me the amount of memory used per second? That would be a bit pointless if you can't tell what the maximum usage of the job was. Unless if that one is showed in the maxvmem value per job. In that case I could get something usable out of it.
            But this still leaves me with the problem of IO and IOW. These values are not only undefined, they do not count NFS IO(W) as far as I can tell. Is there a way to include NFS IO or to get statistics on NFS IO using the sge tools?

            Kind regards,