[Mauiusers] Blocked Jobs not Resuming

Jeff Anderson-Lee jonah at eecs.berkeley.edu
Tue Jul 20 16:16:52 MDT 2010


On 7/20/2010 11:34 AM, Chris Hunter wrote:
> Check maui DEFERTIME & DEFERCOUNT variables. There is a (long) waiting 
> period before blocked jobs are upgraded to idle jobs.
>
> http://www.clustersinc.com/pipermail/mauiusers/2003-October/000898.html
> http://www.clustersinc.com/products/maui/docs/a.fparameters.shtml
>
Hmm. I tweaked the following with my reasoning as comments:

# I'd rather a job sit in the idle queue for days than get blocked
DEFERCOUNT 1000000
# re-evaluate blocked jobs once a minute
DEFERTIME 0:01:00

# prefer that time in the queue is a major factor
QUEUETIMEWEIGHT[0]      10

# look at the prior hour for FairShare factors with steep forgetfulness
FSPOLICY                DEDICATEDPS
FSDEPTH                 6
FSINTERVAL              0:10:00
FSDECAY                 0.50
FSUSERWEIGHT            1

# allow up to 1000 days of wall-clock time
USERCFG[DEFAULT]        MAX.WCLIMIT=86400000
# allow a user 128 processes, backfill upto 256
USERCFG[DEFAULT]        MAXPROC=128,256
# allow a user to use up to 256 nodes (which is more than we have??)
USERCFG[DEFAULT]        MAXNODE=256
# allow even more allocation for the idle queue to prevent jobs getting 
blocked
USERCFG[DEFAULT]        MAXIPROC=2048 MAXIJOB=256 MAXINODE=256
# allow around 50% of the queue to any user before penalizing
USERCFG[DEFAULT]        FSTARGET=50


Am I misunderstanding something?


More information about the mauiusers mailing list