Take a look at the node "Weight" option described in the slurm.conf
man page. The backfill scheduling module does consider job memory
requirements too.
Quoting Ulf Markwardt <[email protected]>:
Dear Slurm developers,
where can I find the rules Slurm uses for the allocation of CPUs and
memory? How can I fine-tune them?
We have nodes with 2, 4, and 8 GB RAM per core. Jobs with low mem
requirements should normally run only on the small nodes. But it
would be a good idea to fill the larger nodes, if the run-time of
the smaller job allows a "backfill" there.
Example: A job allocates 4 cores and 100 GB on a large node, leaving
12 cores and 20 GB free. Now I could fill the node with smaller jobs
as long as they finish before the large one.
How does Slurm handle this? What options do I have for fine-tuning?
Thank you,
Ulf
--
___________________________________________________________________
Dr. Ulf Markwardt
Dresden University of Technology
Center for Information Services and High Performance Computing (ZIH)
01062 Dresden, Germany
Phone: (+49) 351/463-33640 WWW: http://www.tu-dresden.de/zih
