Andromeda is a shared resource, so it is important to be mindful of how your usage may impact other researchers.
- The login nodes are the shared gateway to the cluster, so if they are overwhelmed by compute tasks, it will adversely affect anyone trying to connect to Andromeda. Don’t run any computational job on the login node.
- Compute nodes are a shared resource, avoid arbitrarily using a large portion of the cluster whenever possible; not only will it adversely affect your fair-share priority for later jobs, but it will also make it difficult for other users to submit their workloads.
- Compute nodes allocated to a job should never sit idle
- Requested cores and memory should be the minimum necessary for the job to complete within the time limit
- Make a good-faith effort to make sure that the resources you request are necessary for the job you are running. For instance, if your job only runs on 4 cores, do not request a 64-core node for yourself, the other 60 cores will be blocked from doing useful work by others. This also applies to memory and GPU usage. Try to be as accurate as possible when specifying memory and cores for jobs.
- Temporary/Ephemeral data should be sent to /scratch. For example: when running Gaussian the temporary files are not needed after the job run and do not need snapshots taken.
- Sharing accounts/passwords is a violation of BC policy; All BC IT Policies apply to the cluster
- The HPC cluster is to be used for Research and other sanctioned activities only.
Recommended Job Limits (not enforced):
- 1000 cores per user
- 2 GPUs per user
* Job limits do not apply to dedicated or reserved resources.