Description
A new job scheduler (SLURM) is available for testing on the cluster. It is much faster and more lenient (no pesky walltime limits, for now). A quick overview of commands is provided below (but also consult man pages). For now OAR remains available will likely be removed later this year, once everyone is updated (including TVB). Nb. OAR only employs the first 20 of 28 nodes on the cluster, so 192 of the newest cores (AVX & AVX2) are only accessible via SLURM.
Submit a single job
oarsub "./job.sh a b c"
sbatch ./job.sh a b c
Submit an array of jobs
seq 1 100 > params.txt oarsub --array-param-file params.txt ./job.sh sbatch --array=1-100 ./job.sh (use $SLURM_ARRAY_TASK_ID in job.sh to recover id of array job)
Interactive job
oarsub -I srun --pty /bin/bash
Check usage of cluster
open http://139.124.148.56
- sinfo
- smap
- scontrol show partition
See manpages for more info (man sbatch, man srun, etc)