[GE users] Accounting

Viktor Oudovenko udo at physics.rutgers.edu
Thu Feb 17 17:47:32 GMT 2005


I just want to share my observation and to ask your opinion.
I noticed (from the first glance) that SGE "qacct" does not take into
account that a job runs on many CPU's and records only run time  of  a
program on submission node i.e. MPI job which uses 100 CPU looks the same in
"qacct " a serial job which used only  one CPU, provided they used more or
less  the same time for execution.
I noticed it as in my cluster a user who runs serial jobs has larger
statistics (CPU time and wallclock) than a guy who is heavy MPI runner

Any comments?
If it is true could it be corrected.


To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net

More information about the gridengine-users mailing list