[Mauiusers] jobs not FIFO scheduled

Marc Noguera marc at klingon.uab.es
Tue Jan 27 02:17:27 MST 2009


Hello everyone,
we started using maui as scheduler just some days ago. We use Maui
3.2.16p21 on top of torque 2.3.3 on a fedora core 6 Linux.
Things go pretty well but I am seeing this strange behaviour from time
to time. Some job is keeped in a Q state even if they have resources
available of the queue class, this is already stange. However, some
minutes later when another user sends another job to the same queue,
this one starts running immediately and therefore the FIFO scheme is not
mantained. I am receiving some complains from some users(not the ones
that have their jobs running, of course :)) and, although maui is
working well in general, I would like to solve this. I am probably
missing something in my configuration, and have tried with different
options but this still happens. I have checked priorities using
'diagnose -p' and seen that some jobs are not prioritized accordingly to
their jobid (and therefore the time they have spent on queue)

This is my config file:

SERVERHOST            quanta.uab.es
ADMIN1                mauiuser root
RMCFG[QUANTA.UAB.ES] TYPE=PBS
AMCFG[bank]  TYPE=NONE
RMPOLLINTERVAL        00:00:30
SERVERPORT            42559
SERVERMODE            NORMAL
LOGFILE               maui.log
LOGFILEMAXSIZE        1000000000
LOGLEVEL            2
QUEUETIMEWEIGHT       1
BACKFILLPOLICY       BESTFIT
RESERVATIONPOLICY     NEVER
NODEALLOCATIONPOLICY  MINRESOURCE
DEFERTIME 0
CLASSCFG[parallel]
CLASSCFG[qdynamics6]
CLASSCFG[qdynamics7]
CLASSCFG[qgetab1]
CLASSCFG[qgetab2]
CLASSCFG[qstruct2]
CLASSCFG[qserq1]
CLASSCFG[qserq2]
CLASSCFG[transmet2]
CLASSCFG[transmet3]
CLASSCFG[transmet4]
CLASSCFG[transmet5]
CLASSCFG[transmet1]
CLASSCFG[qstruct1]
CLASSCFG[nahaste]
CLASSCFG[obelix]
CLASSCFG[joanpauii]
CLASSCFG[soca]



Any help is appreciated

Marc


More information about the mauiusers mailing list