Quoting Chi Shin Hsu <[email protected]>:

> Hi,
>
> The slurm default resource policy is exclusive mode for all resource.
>
> Can I use other to control my resource? For example, one node can execute
> 1000 jobs.
>
> Because my tasks  are not use much cpu resource but waste time.
>
> I have tried to use fast schedule  and set 1000 cpus per node.
>
> It works!  But is it the only way to do this?

This is the best way. You could also configure the partition's  
"Shared" parameter, but that would have higher overhead.


> Another question, I can only execute 100 jobs per node, and the remained
> jobs stay in PD state because of resource.

There seems to be a bug preventing more than 100 jobs per node. See  
fix here to increase limit to 10000 jobs:
https://github.com/SchedMD/slurm/commit/952401ec2a56da2b005e6870a0121ccb391736bc

> I can see there are non-allocated cpu resource by scontrol command.
>
>    NodeName=node5 Arch=x86_64 CoresPerSocket=300
>    CPUAlloc=100 CPUErr=0 CPUTot=1200 Features=(null)
>    Gres=(null)
>    NodeAddr=node5 NodeHostName=node5
>    OS=Linux RealMemory=3000 Sockets=4
>    State=MIXED ThreadsPerCore=1 TmpDisk=0 Weight=1
>    BootTime=2012-05-29T22:42:24 SlurmdStartTime=2012-06-02T20:56:19
>    Reason=(null)
>
> I have changed MaxTasksPerNode to 2000 in the configure file, but it's not
> working.
>
> Thanks.
>
> Best regards,
>
> Shin
>

Reply via email to