Re: [Condor-users] distributing job load

[Jaime Frey <jfrey@xxxxxxxxxxx>]

On Nov 25, 2011, at 12:53 AM, Tom Melendez wrote:

> Thanks so much for the help so far.  I appreciate the quick responses
> people have given me and they've right-to-the-point and again, quite
> helpful.  Thanks.
> 
> I'm running my sample job, which generates 45 tasks.  I'm noticing
> that my 45 tasks are consuming complete (multi-core) machines.  I'd
> like to distribute this a little more evenly across the nodes that I
> have available.  I'm assuming this gets me into the topic of setting
> policies for my cluster.  Am I correct about this?
> 
> As this is a cluster available for my own use, I'm just looking for a
> round-robin type distribution, so I can get an even workload across my
> machines.  Ideas on how I can accomplish this?


One way to do this is to rank the machines based on SlotID. 
You can do this in the submit file:
   rank = -TARGET.SlotID
Or you can do this in the negotiation's configuration for all jobs:
   NEGOTIATOR_POST_JOB_RANK = -SlotID

This will cause jobs to prefer slot 1 of each machine over slot 2 or higher of any machine, and should spread your jobs out across the machines more evenly.

Thanks and regards,
Jaime Frey
UW-Madison Condor Team