Difference between revisions of "Accounting"
| Line 1: | Line 1: | ||
Torque keeps records of job wall and CPU times on the torque server, directory | Torque keeps records of job wall and CPU times on the torque server, directory | ||
| + | /var/spool/pbs/server_priv/accounting | ||
| + | What is important to our users is not the time used, but the total work performed, which for each job consists of the actual times (cpu and wall) multiplied by the power rating of the machine running the job. | ||
| + | The relevant conversion factors are contained in the file | ||
| + | /etc/pbsaccdb.conf | ||
| + | on the torque server. This file is controlled by Quattor, so to make changes (for example if a new worker-node class is added), make the changes in the quattor profile and not on the torque server. | ||
| + | Once per day, a cron job runs on the torque server: | ||
| + | /etc/cron.d/accounting.ncm-cron.cron | ||
| + | This processes the torque accounting file for the previous day. If one needs to re-run the processing for any reason, one can do it by hand via | ||
| + | /share/acct/bin/pbsaccdb.pl <pbs_accounting_file> [<pbs_accounting_file> ...] | ||
| + | assuming your CWD is the torque accounting directory. | ||
| + | ''Need information on the rest of the chain here, for example where these data are stored, and how the APEL part of the chain works. Also a human-readable explanation of the dreaded normalization is needed.'' | ||
Revision as of 14:44, 11 July 2008
Torque keeps records of job wall and CPU times on the torque server, directory
/var/spool/pbs/server_priv/accounting
What is important to our users is not the time used, but the total work performed, which for each job consists of the actual times (cpu and wall) multiplied by the power rating of the machine running the job. The relevant conversion factors are contained in the file
/etc/pbsaccdb.conf
on the torque server. This file is controlled by Quattor, so to make changes (for example if a new worker-node class is added), make the changes in the quattor profile and not on the torque server. Once per day, a cron job runs on the torque server:
/etc/cron.d/accounting.ncm-cron.cron
This processes the torque accounting file for the previous day. If one needs to re-run the processing for any reason, one can do it by hand via
/share/acct/bin/pbsaccdb.pl <pbs_accounting_file> [<pbs_accounting_file> ...]
assuming your CWD is the torque accounting directory. Need information on the rest of the chain here, for example where these data are stored, and how the APEL part of the chain works. Also a human-readable explanation of the dreaded normalization is needed.
