[slurm-users] Slurm 21.08 / split node in to two partitions

Muhammad Hassan Khan email.hassan12 at gmail.com
Thu Mar 31 04:23:06 UTC 2022


How are these nodes defined in slurm.conf file? Using expressions or each
node is defined in a separate line? One way to do this is to define gpu
nodes separately in node configuration in slurm.conf file and define a
partition for these nodes only:

NodeName=gpu[0-3] Gres=gpu:4 cpus=64 ...
partitionName=gpupar Nodes=gpu[0-3] ...

This is just a sample. Actual configurations will depend on node
specifications as well.

On Thu, Mar 31, 2022 at 6:57 AM pankaj d <purveshp0507 at gmail.com> wrote:

> Hi,
>
> We have 8 fat nodes with 4 GPUs and 128 CPUs with 512 GB RAM each. We want
> to split the 8 nodes in to 2 partitions so that one partition takes 4 GPUs
> and 64 cores with 256 GB RM (gpu partition) and other partition takes same
> node with remaining 64 cores (in cpu partition with remaining 256 GB RAM),
> so a node with nodename n01 will be present in gpupar and cpupar with
> different configuration like gpupart will have n01 listed as 4 gpus, 64
> cores, 256 GB RAM and cpupar will have n01 with remaining 64 cores and 256
> GB RAM . How to achieve this?
>
> Regards,
>
> Purvesh
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20220331/cfe1a7e0/attachment.htm>


More information about the slurm-users mailing list