The Slurm job scheduler uses partitions to organize job submissions from researchers. Andromeda 2 has partitions that structured to separated jobs by the amount of time they require to complete, preferencing shorter running jobs to facilitate the fast paced demands of BC’s research community.
The “partition priority” for each partition listed below in on of many factors the Slurm job scheduler uses to determine the priority of a submitted job in the queue. Some partitions have a small number of nodes that their jobs exclusively can be run on, while the majority of the compute nodes are accessible to all of the partitions. Which ensures ensures jobs that are next in the queue have the greatest opportunity to be launched as soon as running jobs in any of the other partitions are relinquish their resources when they finish.
Jobs requiring few resources are inherently easier for Slurm to schedule. And long running jobs make it a bit more tricky for Slurm to find resources to allocate to jobs quickly. So, Andromeda’s partition scheme is tailored to slightly incentivize larger, short running jobs. Not all jobs can take advantage of multiple nodes, so small and large jobs have equal access on Andromeda. Researcher’s are encouraged to use whatever resources they might need as efficiently as possible, so they can submit their jobs to shortest timed partition to get the highest priority in the queue.
Job priority is largely based on the researcher’s fairshare value, which goes down relative to their use of the cluster within a two week period. The partition priorities are not large enough to overcome a low fairshare value. There are other Slurm features that can have a much greater effect on how quickly a low priority job starts, such as the “backfill” plugin and the “age” factor. So, don’t fret if your jobs do need to run for 5 days.
If you find you’re in urgent need of cluster resources to meet a deadline, please let us know as soon as possible by submitting a request for assistance here: Research Help. We can’t guarantee access to compute resources, but will do our best to help in whatever way we can. The sooner we receive a request, the more likely it is a solution can be found. Don’t delay!
Andromeda 2 Slurm Partitions
- interactive
- 12 hour time limit
- Highest partition priority
- Max of 2 running jobs per user
- Max of 16 cores per user across all running jobs
- Max of 64GB of RAM per user across all running jobs
- short
- 12 hour time limit
- Next to highest partition priority
- medium
- 48 hour time limit
- Moderate partition priority
- long
- 120 hour time limit
- Lowest partition priority