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

[Condor-users] RANK and job placement

I have a condor pool with ~250 batch slots available.  On half of the
batch slots I have set:

RANK = (UserA =?= TRUE)

...and on the other half of the batch slots I have:

RANK = (UserB =?= TRUE)

This will allow both users to make use of free batch slots on the
cluster, while guaranteeing that each user has access to 50% of the
cluster resources.

This has worked out pretty well for us so far.  But I've noticed some
interesting behavior that I was hoping someone could help me tune.
Occasionally I notice that UserA has ~20 jobs running, with the rest of
the batch slots empty.  These 20 jobs are scattered across slots where
the user has RANK priority and slots where the user does not have RANK
priority.  Next, UserB comes along and submits a bazillion jobs.  This
causes some of UserA's jobs to get evicted and restarted because some of
UserA's jobs are running on batch slots where UserB has rank priority.

When UserA submitted his 20 jobs, why were they not directed to batch
slots where UserA had rank priority?  How can I get UserA's jobs to
prefer those batch slots while still allowing him opportunistic access
to the rest of the cluster?