Hi,
I'm trying to get maui's suspend feature to work (migrating from torque's
native pbs_sched); the idea is to suspend long running jobs when short ones
are submitted, either manually thru the users or by the scheduler after a
certain "starving" time.
Torque has 2 queues, "default" and "short".
"short" automatically defines 2:00 as walltime limit, "default" has no
walltime limit set:
# default
set queue default resources_default.nodes = 1:ppn=1
# short
set queue short resources_max.walltime = 02:00:00
set queue short resources_default.nodes = 1:ppn=1
set queue short resources_default.walltime = 02:00:00
According to the docs I've set up maui.cfg:
PREEMPTPOLICY SUSPEND
CLASSCFG[default] FLAGS=PREEMPTEE
CLASSCFG[short] FLAGS=PREEMPTOR MAXNODE=8,12
QOSCFG[short] PRIORITY=100 QFLAGS=PREEMPTOR
QOSCFG[default] PRIORITY=500 QFLAGS=PREEMPTEE
At the moment all nodes are in use via "default" jobs.
I submitted a couple to short, then I tried manually to suspend a "default"
job via
#> mjobctl -s <JOBID>
That worked, BUT no job from "short" is started; instead, after the next
maui scheduling run, the suspended job goes back to state "Running".
Any suggestions?
Cheers,
Ronny
_______________________________________________
mauiusers mailing list
[email protected]
http://www.supercluster.org/mailman/listinfo/mauiusers
