We have been running maui on a cluster for some time and are trying to
get it to also run on our 128 processor Altix (itanium2) shared memory
machine, but the standing reservations are not being honoured. 

In the first instance I am configuring it on a 16p partition, with a
standing reservation to set aside 3 processors for short jobs:

 SRCFG[hour6] TASKCOUNT=3 MAXTIME=6:00:00
 SRCFG[hour6] RESOURCES=PROC:1
 SRCFG[hour6] FLAGS=DEDICATEDRESOURCE

/usr/local/maui # showres
-snip-
hour6.0.0           User -    00:00:00     9:17:24     9:17:24    1/3
Wed Jul 12 14:42:36
hour6.1.0           User -     9:17:24  1:09:17:24  1:00:00:00    1/3
Thu Jul 13 00:00:00

When I submit a 5hr, 15p job it starts immediately on the scheduling
cycle triggered by job submission (qsub -I -l
nodes=1:ppn=15,walltime=5:00:00), with the hour6 reservation retained.

328                  Job R    00:00:00     5:00:00     5:00:00    1/15
Wed Jul 12 14:54:31
hour6.0.0           User -    00:00:00     9:05:29     9:05:29    1/3
Wed Jul 12 14:54:31
hour6.1.0           User -     9:05:29  1:09:05:29  1:00:00:00    1/3
Thu Jul 13 00:00:00

Subsequently the hour6 reservation is reduced:

328                  Job R   -00:00:31     4:59:29     5:00:00    1/15
Wed Jul 12 14:54:31
hour6.0.0           User -    00:00:00     9:04:58     9:04:58    1/1
Wed Jul 12 14:55:02
hour6.1.0           User -     9:04:58  1:09:04:58  1:00:00:00    1/3
Thu Jul 13 00:00:00

However, when I submit a 7hr, 15p job, on the first scheduling cycle the
jobs gets a reservation in the future:

326                  Job I  1:09:12:30  1:16:12:30     7:00:00    1/15
Fri Jul 14 00:00:00
hour6.0.0           User -    00:00:00     9:12:30     9:12:30    1/3
Wed Jul 12 14:47:30
hour6.1.0           User -     9:12:30  1:09:12:30  1:00:00:00    1/3
Thu Jul 13 00:00:00

But on the next cycle the job is started:

326                  Job R    00:00:00     7:00:00     7:00:00    1/15
Wed Jul 12 14:48:01
hour6.0.0           User -    00:00:00     9:11:59     9:11:59    1/3
Wed Jul 12 14:48:01
hour6.1.0           User -     9:11:59  1:09:11:59  1:00:00:00    1/3
Thu Jul 13 00:00:00

And on a later scheduling cycle the hour6 reservation has the number of
tasks/processors reduced to 1:

326                  Job R   -00:02:35     6:57:25     7:00:00    1/15
Wed Jul 12 14:48:01
hour6.0.0           User -    00:00:00     9:09:24     9:09:24    1/1
Wed Jul 12 14:50:36
hour6.1.0           User -     9:09:24  1:09:09:24  1:00:00:00    1/3
Thu Jul 13 00:00:00



There seem to be at least two problems.
1) The 7hr job is scheduled regardless of the reservation (at least on
the 2nd scheduling cycle)
2) The reservation is subsequently adjusted/reduced, even in the case of
the 'allowable' 5hr job.

Is anybody else experiencing these problems, and does anyone have a
solution?

We are running the latest released versions of torque and maui.

Thanks,

Gareth Williams, CSIRO HPSC
_______________________________________________
mauiusers mailing list
[email protected]
http://www.supercluster.org/mailman/listinfo/mauiusers

Reply via email to