Impact jobs can become quite a lot (even if mostly short, and they can be grouped with a few per job), it's more efficient (in terms of load on the scheduler) to submit a single "job cluster" to slurm than each as a separate job (similary to task arrays with HTCondor).
adewit
commented
3 years ago
Impact jobs can become quite a lot (even if mostly short, and they can be grouped with a few per job), it's more efficient (in terms of load on the scheduler) to submit a single "job cluster" to slurm than each as a separate job (similary to task arrays with HTCondor).