[slurm-users] Oversubscribe until 100% load?

[Holtgrewe, Manuel]

Hi,

I have some trouble understanding the "Oversubscribe" setting completely. What I would like is to oversubscribe nodes to increase overall throughput.

- Is there a way to oversubscribe by a certain fraction, e.g. +20% or +50%?
- Is there a way to stop if a node reaches 100% "Load"?

Is there any good documentation available online that describes how to "carefully oversubscribe" a cluster?

Our users have pretty mixed workloads, e.g., with high parallelism in the first 50% of the wall-clock time, then parts with mixed parallelism. Of course, we should educate our users better, but in some cases, it's very hard to improve because of the software used or workloads that cycle between being I/O and compute bound.

Thank you,
Manuel

--
Dr. Manuel Holtgrewe, Dipl.-Inform.
Bioinformatician
Core Unit Bioinformatics – CUBI
Berlin Institute of Health / Max Delbrück Center for Molecular Medicine in the Helmholtz Association / Charité – Universitätsmedizin Berlin

Visiting Address: Invalidenstr. 80, 3rd Floor, Room 03 028, 10117 Berlin
Postal Address: Chariteplatz 1, 10117 Berlin

E-Mail: manuel.holtgrewe at bihealth.de
Phone: +49 30 450 543 607
Fax: +49 30 450 7 543 901
Web: cubi.bihealth.org  www.bihealth.org  www.mdc-berlin.de  www.charite.de
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20200611/01acae1f/attachment.htm>