Just to close the loop, I found the "CR_LLN" setting to be used in
SelectTypeParamters, which achieved the goal we were after.

On Wed, Dec 3, 2014 at 9:07 AM, David Lipowitz <[email protected]>
wrote:

>  Hello slurm-dev,
>
> I working on recreating some historical workload from our legacy scheduler
> to see how Slurm would handle the same.  The jobs we submitted completed
> successfully, but I noticed they weren't equally distributed across our
> compute nodes:
>
> mysql> select nodelist, count(*) from cluster_job_table where id_job
> between 45470 and 70987 group by nodelist;
> +---------------+----------+
> | nodelist      | count(*) |
> +---------------+----------+
> | cloudlg016185 |     8084 |
> | cloudlg016193 |     8768 |
> | cloudlg016198 |     5864 |
> | cloudlg016199 |     1783 |
> | cloudlg016207 |      157 |
> | cloudlg016208 |      204 |
> | cloudlg016209 |      174 |
> | cloudlg016210 |      484 |
> +---------------+----------+
> 8 rows in set (1.17 sec)
>
>
> Is there a setting I can change so the distribution is more even?  I've
> attached our slurm.conf for reference.
>
> Thanks in advance for any assistance.
>
>
> Cheers,
> Dave
>



-- 
David Lipowitz
720-684-2403
[email protected]

Reply via email to