[slurm-users] How to increase the cpu and gpu usage in our Slurm cluster?

[feng lu]

Hi All

Our Slurm CPU and GPU cluster have 22 and 150 machines and 1400 core ( open
hyper-threading）, 900 GPU card。 Now the cpu occupancy is up to 90%, but the
usage is about 25% percent. How to crease the cpu usage in our cluster? I
can think following methods

1. Virtualize our cpu core using docker or other similarly tech
2. Optimize resource apply strategy for our tasks
3. Optimize our code and computing framework

Can you show me some viable approach for our Slurm cluster?

Thanks

On Thu, May 7, 2020 at 8:29 PM feng lu <amuseme.lu at gmail.com> wrote:

> Hi All
>
> Our Slurm CPU and GPU cluster have 22 and 150 machines and 1400 core (
> open hyper-threading）, 900 GPU card。 Now the cpu occupancy is up to 90%,
> but the usage is about 25% percent. How to crease the cpu usage in our
> cluster? I can think following methods
>
> 1. Virtualize our cpu core using docker or other similarly tech
> 2. Optimize resource apply strategy for our tasks
> 3. Optimize our code and computing framework
>
> Can you show me some viable approach for our Slurm cluster?
>
> Thanks
>
> --
> Don't Grow Old, Grow Up... :-)
>


-- 
Don't Grow Old, Grow Up... :-)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20200509/7a82367b/attachment-0001.htm>