Contents
Migration from LSF to SLURM
Basic commands: |
LSF |
SLURM |
| job submission | bsub < jobscript | sbatch jobscript (no "<") |
| job lookup | bjobs | squeue -u <user_list> |
| job deletion | bkill job_id | scancel job_id |
| script prefix | #BSUB | #SBATCH |
Options and Parameters for Batch Scripts
Optional Parameters are placed in [].
Functionality |
LSF |
SLURM |
| run time | #BSUB -W [hh:]mm |
#SBATCH -t [[dd-]hh:]mm[:ss] #SBATCH -t 0:30 will be 30 sec. |
| memory (4GB) on node | #BSUB -R "rusage[memory=4096]" | #SBATCH --mem=4096 |
| selecting a queue | #BSUB -q <queue> | #SBATCH -p <partition> |
| selecting a project | #BSUB -p <project> | #SBATCH --account=<project> |
| specifying a job name | #BSUB -J <job name> | #SBATCH -J <job name> |
| mail notification | automatic at end | #SBATCH --mail-type=END |
| selecting processor type |
#BSUB -R "select[model==XEON_E5_2670]" #BSUB -R "same[model]" |
#SBATCH --constraint=XEON_E5_2670 not necessary. Parallel jobs use always homogeneous hardware. |
| starting at a certain time |
#SBATCH --begin hh:mm:ss #SBATCH --begin yyyy-mm-dd[Thh:mm[:ss]] | |
| naming job output/error file |
#BSUB -o output.%J #BSUB -e error.%J |
#SBATCH -o output.%j #SBATCH -e error.%j |
All SLURM batch scripts should begin with a shell specification like #!/bin/bash