Hello everyone,
we have a strange problem with maui: with backfill enabled,
reservations for high-priority four-node jobs are slipping forward in
time; thus, low-priority two-node jobs manage to occupy free nodes,
effectively bypassing the queue.
A reservation spanning both free and occupied nodes would be created,
keeping the free nodes free. Then, the other nodes that are part of
the reservation would become available, and the reservation would be
moved to a later point in time. Consequently, low-priority jobs would
be backfilled into the now-freed nodes.
At first, I could not make heads or tails of this behaviour, but
looking through maui.log, I suspect something may be amiss with our
partitions: it seems the reservation for job 2645 is created within
the default partition rather than one of IBA, IBB, IBC.
However, I am at a loss as to how to change the configuration.
12/16 18:01:17
MPBSJobUpdate(2645,2645.master1.beowulf.cluster,TaskList,0)
12/16 18:01:17 INFO: 192 feasible tasks found for job 2645:0 in
partition DEFAULT (32 Needed)
12/16 18:01:17 MJobPReserve(2645,DEFAULT,ResCount,ResCountRej)
12/16 18:01:17 INFO: 192 feasible tasks found for job 2645:0 in
partition DEFAULT (32 Needed)
12/16 18:01:17 INFO: 192 feasible tasks found for job 2645:0 in
partition IBA (32 Needed)
12/16 18:01:17 INFO: 192 feasible tasks found for job 2645:0 in
partition IBB (32 Needed)
12/16 18:01:17 INFO: 192 feasible tasks found for job 2645:0 in
partition DEFAULT (32 Needed)
12/16 18:01:17 INFO: located resources for 32 tasks (32) in best
partition DEFAULT for job 2645 at time 00:07:26
12/16 18:01:17 INFO: tasks located for job 2645: 32 of 32
required (0 feasible)
12/16 18:01:17 INFO: job '2645' reserved 32 tasks (partition
DEFAULT) to start in 00:07:26 on Tue Dec 16 18:08:43
12/16 18:01:19
MPBSJobUpdate(2645,2645.master1.beowulf.cluster,TaskList,0)
12/16 18:01:19 INFO: 192 feasible tasks found for job 2645:0 in
partition DEFAULT (32 Needed)
12/16 18:01:19 MJobPReserve(2645,DEFAULT,ResCount,ResCountRej)
12/16 18:01:19 INFO: 192 feasible tasks found for job 2645:0 in
partition DEFAULT (32 Needed)
12/16 18:01:19 INFO: 192 feasible tasks found for job 2645:0 in
partition IBA (32 Needed)
12/16 18:01:19 INFO: 192 feasible tasks found for job 2645:0 in
partition IBB (32 Needed)
12/16 18:01:19 INFO: 192 feasible tasks found for job 2645:0 in
partition IBB (32 Needed)
12/16 18:01:19 INFO: located resources for 32 tasks (32) in best
partition IBB for job 2645 at time 00:12:43
12/16 18:01:19 INFO: tasks located for job 2645: 32 of 32
required (0 feasible)
12/16 18:01:19 INFO: job '2645' reserved 32 tasks (partition IBB)
to start in 00:12:43 on Tue Dec 16 18:14:02
Looking forward to any suggestions,
A.
--
Ansgar Esztermann
DV-Systemadministration
Max-Planck-Institut für biophysikalische Chemie, Abteilung 105
_______________________________________________
mauiusers mailing list
[email protected]
http://www.supercluster.org/mailman/listinfo/mauiusers
