[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] About negotiation

> > This is part of the "high throughput" portion of Condor. A claim on
> > startd remains in place until it: a) runs out of jobs to process
> > the cluster; or b) gets preempted by another claim. As long as
> > no one with a lower user priority value in the system it's much more
> > efficient to keep cycling through the jobs from the current cluster
> > being executed than re-negotiate because you don't have to tear down
> > setup the shadow again.
> OK--now I am really confused.  Other E-mails on other threads have
> said that the schedd will keep the claim on the startd as long
> as it has any jobs for that user (if PREEMPTION_REQUIREMENTS = FALSE
> and RANK=0).  Now
> you are saying it is just
> if the jobs are in the same cluster.  Which is it?

I'm 90% certain the claim expires when there a no more jobs from the
same *cluster*. The Condor team can give you the 100% confidence answer
on that one though. If you're using auto-clustering so that clusters are
built dynamically at negotiation time, grouping like jobs from different
physical clusters into the same auto-cluster, this *might* change the
observed behaviour. I'm not sure how auto-clusters and the startd

- Ian