Quick reference
Useful hints and short information on COSMOS resources and the SLURM batch system.
Installed software
To list all software available through the module system:
To check prerequisites for a specific package:
To see currently loaded modules:
To load a module:
To unload a module:
Resource allocation
Number of cores
Specify nodes and tasks per node in your job script:
COSMOS has 48 cores per node. To allocate 192 cores across 4 nodes:
Memory per core
COSMOS nodes have 256 GB of memory. The default allocation is 5300 MB per core, leaving headroom for the operating system. Requesting more than 5300 MB per core will leave some cores idle — your account is charged for those idle cores too.
File systems
Home directory
Your home directory on COSMOS is:
Also accessible as $HOME.
Global working directory
Use your home directory for job submission. The old /lunarc/nobackup/users directory has been decommissioned.
Local working directory
Each compute node has a local scratch disk available during a job at $SNIC_TMP (also $TMPDIR). It is deleted when the job ends.
If a job is terminated early (e.g. walltime exceeded), files on the local disk are lost. To preserve specific files, list them in $SNIC_TMP/slurm_save_files — paths should be relative to $SNIC_TMP, separated by spaces or newlines. These files will be copied to the submission directory if at all possible.
Quotas
Disk quotas are set per user and filesystem. Your quota status is shown at login, or run:
Quota increases are available on request.
Test queue
On COSMOS, short tests (maximum 1 hour, at most 2 nodes) can be given elevated priority:
Do not submit long series of jobs to the test queue.
Author: (LUNARC)
Last Updated: 2024-06-18