[Condor-users] sheduling policy with large no of jobs

["Dr Ian C. Smith" <i.c.smith@xxxxxxxxxxxxxxx>]

Hi,

I've another quick question about scheduling large
numbers of jobs although I imagine on past form it
may not have a quick answer !

Our Windows pool is currently running in cycle-stealing/
SETI@home type mode. If a local user logs in on a pool
PC then any condor job on it will be vacated. We
have a user submitting big clusters of jobs (on the order
of 100 * pool size). There is a wide variety in the
execution time of the individual jobs so some are unlikely
to reach completion.

The question is what is the policy for running jobs that
have previously been kicked out of the pool. Do they
go to the back of the queue ( a sort of "first out
last in" ) or does the schedd pick jobs at random for a cluster.
Or some other policy ?

Ideally I'd like to see the shorter jobs completing quickly leaving
just the long ones behind. That way the long ones (which are
unlikely to ever complete) can be identified easily and removed.

regards,

-ian.


-----------------------------------
Dr Ian C. Smith,
e-Science team,
University of Liverpool
Computing Services Department