Brian Smith <[email protected]> writes: > max_reservation seems to set the limit on the number of complex > attributes that can be reserved by the highest priority jobs for the > purposes of preventing starvation by smaller jobs contending for fewer > resources. So, if I set max_reservations to 128 and someone submits a > 256 processor job, its very possible that the job will be starved on a > highly utilized system.
What version is behaving like that? I have max_reservations 30 and there are currently these reservations (and, I hope, a patient submitter of the long-running 128-core jobs): # qsched -a job-ID user start end slots h_vmem -------- -------- ------------------- ------------------- ----- ------ 154082.1 ******** 08/25/2012 09:46:19 08/30/2012 09:51:19 32 2000M 154389.1 ******** 08/27/2012 17:45:53 09/04/2012 17:49:53 64 4000M 154585.1 ******** 08/28/2012 00:36:22 09/01/2012 00:41:22 32 4000M 154486.1 ******** 08/28/2012 05:05:26 09/02/2012 05:10:26 36 2000M 153880.1 ******** 08/30/2012 09:51:19 09/04/2012 09:56:19 128 2000M 153881.1 ******** 09/04/2012 09:56:19 09/09/2012 10:01:19 128 2000M 153884.1 ******** 09/09/2012 10:01:19 09/14/2012 10:06:19 128 2000M 153885.1 ******** 09/14/2012 10:06:19 09/19/2012 10:11:19 128 2000M 153886.1 ******** 09/19/2012 10:11:19 09/24/2012 10:16:19 128 2000M 153887.1 ******** 09/24/2012 10:16:19 09/29/2012 10:21:19 128 2000M 153888.1 ******** 09/29/2012 10:21:19 10/04/2012 10:26:19 128 2000M 153889.1 ******** 10/04/2012 10:26:19 10/09/2012 10:31:19 128 2000M 153891.1 ******** 10/09/2012 10:31:19 10/14/2012 10:36:19 128 2000M -- Community Grid Engine: http://arc.liv.ac.uk/SGE/ _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
