[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] about priorities
- Date: Thu, 09 Sep 2004 13:41:25 +0200
- From: Sven-Arne Reinemo <svenar@xxxxxxxxx>
- Subject: Re: [Condor-users] about priorities
On Thu, 2004-08-05 at 04:43, Erik Paulson wrote:
> On Wed, Aug 04, 2004 at 06:16:45PM -0500, David A. Kotz wrote:
> > I think what you want to do is to set the following in your
> > condor_config:
> > SUSPEND = FALSE
> > PREEMPT = FALSE
> > KILL = FALSE
> That will help prevent the local machine from evicting the job (you'll also
> probably want have START = TRUE, or some such)
> Those expresssions have nothing to do with preemption for priority
> purposes, though. Condor can "preempt" a job for two reasons - one,
> because the local conditions on the machine force us to get rid of a job,
> or two, because there is a user with a higher priority who wants to run a
> job. That preemption is controlled by PREEMPTION_REQUIREMENTS. It is a
> pool-wide setting that is evaluated only when Condor has a higher
> priority user who wants to run a job. It is tested against each machine
> in the pool to see if the job currently running on that machine
> can be preempted.
> To prevent priority preemption, set:
> PREEMPTION_REQUIREMENTS = FALSE
> in the config file of your central manager.
Is it possible to do something like this on just a subset of nodes?
In our pool we have problems with long running jobs in the Java
universe. They never finish since they are preempted by users with
better priority and then have to start all over when they are
rescheduled (then they are preempted by users with better priority...
etc.). Since our pool is a mix of workstations and dedicated computing
resources turning of preemption on all nodes does not sound like a good
---- GnuPG public key - http://home.ifi.uio.no/~svenar/gpg.asc ----
"There are only 10 kinds of people in this world; those who know
binary and those who don't."