Once on Andromeda 2 you’ll find a mostly similar setup to Andromeda 1, with a few key differences:
- Storage quotas
- Your home directory (/home/<login>) has a 50G initial quota
- Intended for storing personal data only
- Thirty days of nightly snapshots
- The project directory (/projects/<project_name>) has a 5TB initial quota
- Intended for storing research related data that is shared with the group
- 30 days of nightly snapshots
- The scratch directory (/scratch/<login>) has a 5TB initial quota
- Intended for storing ephemeral data that can be easily replaced and does not need to be shared, or is only needed for a short time (e.g. While a job is running.)
- Your home directory (/home/<login>) has a 50G initial quota
- Open OnDemand (ood.bc.edu)
- Replaces NoMachine on top of providing a significant amount of addition functionality
- OOD Remote Desktop Sessions have a maximum time limit of 12 hours
- The session itself is run inside a Slurm job on one of the compute nodes
- Session are only allocated 2 cores and 4G of RAM
- If additional resources are needed a batch job can be submitted or the salloc command can be used to work interactively
- Login node (a002.bc.edu) resources
- 8 cores and 96G of RAM
- Intended for file transfers, SSH access and job submissions
- Access to module packages is very limited, compared to the compute nodes
- Will soon be monitored by Arbiter to ensure login node resources are not used improperly
- Compute nodes
- Currently there are 19 CPU nodes available
- CPU core reservations per node
- Ensures jobs cannot starve the OS, Slurm or the Weka storage agent
- 2 cores for the OS and Slurm
- 2 cores for Weka
- RAM reservations per node
- Ensures jobs cannot starve the OS, Slurm or Weka
- 5G per compute node
- Minimal OS
- As few packages have been installed to the local disk as possible
- vm.swappiness set to 1
- Avoids, as much as possible, the use of 400G swap area when physical RAM is available
- Slurm Partitions
- All compute nodes are currently available to the primary partitions
- As more compute nodes are added each partition will have a subset of compute nodes dedicated to them, based on usage
- “Interactive” partition only has one node, which is reserved for OOD Desktop sessions, for now
- Primary Partitions
- short – 12 hour time limit (highest partition priority)
- medium – 48 hour time limit (moderate partition priority
- long – 120 hour time limit (lowest partition priority)
- All compute nodes are currently available to the primary partitions