It's configurable on a per-partition basis. Excerpt from slurm.conf
man page below:
CR_LLN
Schedule resources to jobs on the least loaded nodes (based upon
the number of idle CPUs). This is generally only recommended for an
environment with serial jobs as idle resources will tend to be highly
fragmented, resulting in parallel jobs being distributed across many
nodes. Also see the partition configuration parameter LLN use the
least loaded nodes in selected partitions.
Quoting Franco Broi <[email protected]>:
We submit jobs with varying Gres counts based on the type of job being
run, I'm not sure a plugin would help make this any easier.
Changing the subject slightly, how does the load balancing mentioned in
the release notes work? Is it only designed to work with arrays?
On Wed, 2014-04-09 at 19:18 -0700, [email protected] wrote:
You can use a job_submit plugin to automatically enter/edit a GRES
value and avoid forcing users to enter it (assuming it applies to all
jobs, e.g. "--gres=jobs").
Quoting Franco Broi <[email protected]>:
> Yep that's how we do it, note that you need to specify Gres in three
> places:
>
> in slurm.conf
>
> GresTypes=Mygres
> NodeName=.... Gres=Mygres
>
> and in gres.conf
>
> Name=Mygres Count=2
>
>
> And when you submit a job, use sbatch --gres=Mygres
>
> On Wed, 2014-04-09 at 10:30 -0700, [email protected] wrote:
>> The simplest solution would probably be to define a generic resource
>> (GRES) of "jobs", two per node, and then in a job_submit plugin, set
>> every job to require one GRES of type "jobs" (at least that's what we
>> did on some Cray systems). More information here:
>>
>> http://slurm.schedmd.com/gres.html
>> http://slurm.schedmd.com/job_submit_plugins.html
>>
>>
>> Quoting "Ethier, Michael" <[email protected]>:
>>
>> > Hello,
>> >
>> >
>> >
>> > We have a user asking if there is a way to tell slurm to only
schedule 2
>> > jobs per node at any one time. Is there an easy way to do this ?
>> >
>> >
>> >
>> > Thanks,
>> > Mike
>>
