Are there any tools I can use to optimize my workflows before submitting them to run on RMACC? Similarly, is there a queue with a short wall time (and ideally little to no wait) where I can test jobs before submitting them to a queue with a longer wall time?
Slurmtools is a collection of helper scripts for everyday use of the slurm job scheduler, one of my favorite tools is the “seff” script that gives efficiency information about a given job such as CPU time utilize, CPU efficiency, memory utilized, and memory efficiency. A complete list of the scripts available with the Slurmtools module can be found here: Monitoring Resources — Research Computing University of Colorado Boulder documentation
Example:
$ module load slurmtools
$ seff 1234567
Job ID: 1234567
Cluster: summit
User/Group: ralphie/ralphiegrp
State: COMPLETED (exit code 0)
Nodes: 1
Cores per node: 24
CPU Utilized: 04:04:05
CPU Efficiency: 92.18% of 04:24:48 core-walltime
Job Wall-clock time: 00:11:02
Memory Utilized: 163.49 MB
Memory Efficiency: 0.14% of 113.62 GB
There are also some testing queues that can be used for testing on RMACC Summit which give you access to shorter wait times (but note they also have less resources available).
The following partitions give you access to short turn around jobs on the Summit cluster:
- shas-testing (general CPU node)
- shas-interactive (interactive job)
- sgpu-testing (GPU node)