[torqueusers] Torque job accounting: figuring out run duration

Troy Baer tbaer at utk.edu
Thu Feb 7 15:52:50 MST 2013


On Thu, 2013-02-07 at 22:44 +0000, Nick Lindberg wrote:

> Per those parameters, I am familiar with the job information being
> stored in:
> 
> 
> /var/spool/torque/server_priv/Accounting/ 
> 
> 
> The basic format is:
> 
> 
> 01/17/2013 17:34:10;S;38254.hpc01;user=nlindberg group=nlindberg
> jobname=Job175 queue=batch ctime=1357945580 qtime=1357945580
> etime=1357945580 start=1358465650 owner=nlindberg exec_host={compute
> node names} Resource_List.neednodes=1 Resource_List.nodect=1
> Resource_List.nodes=1 Resource_List.procs=232
> Resource_List.walltime=240:00:00

Look for records in your accounting logs that have an "E" as the second field:

01/17/2013 17:34:10;E;38254.hpc01;user=nlindberg [...]

Those are job end records.  In those, you should see fields like
resources_used.walltime and resources_used.mem.

The following URLs contain Perl examples of how to parse PBS accounting
records:

http://svn.nics.tennessee.edu/repos/pbstools/trunk/sbin/jobstats
http://svn.nics.tennessee.edu/repos/pbstools/trunk/sbin/job-db-update

Hope this helps,
	--Troy
-- 
Troy Baer, Senior HPC System Administrator
National Institute for Computational Sciences, University of Tennessee
http://www.nics.tennessee.edu/
Phone:  865-241-4233




More information about the torqueusers mailing list