Nodes and what you can do there¶
The login nodes¶
These nodes are the entry point to mistral. You reach them via
ssh mistral.dkrz.de. They are mainly for editing text files, starting jobs, looking at an image / …
Many users have to share the memory and disk bandwidth, so please be considerate
On these nodes you can do whatever, but that also means the whole node might be crashed by your neighbors escalating python script or such. Use
ssh mistralpp.dkrz.de to log in. Please behave considerately and don’t be the neighbor that crashes the wohle node.
The batch queues¶
You can use the standard batch queues of mistral for scripts (see https://www.dkrz.de/up/systems/mistral/running-jobs/example-batch-scripts for instructions) and for interactive processing. Each queue has a different specification and limitations, as described at https://www.dkrz.de/up/systems/mistral/running-jobs/partitions-and-limits .
For the interactive use, you can do something like
salloc -p prepost -A xz0123 -n 1 -t 12:00:00 -- /bin/bash -c 'ssh -X $SLURM_JOB_NODELIST'
If you need more resources, getting your own node might be an option:
salloc --exclusive --mem 0 -N 1 -p compute,compute2 -A YOUR_PROJECT_ID -t 8:00:00 -- /bin/bash -c 'ssh -X $SLURM_JOB_NODELIST'
If necessary, this can be combined with
--constraint=256G to get nodes with more memory (those are scarce resources so be kind to others…) Note that this needs the
--mem 0 to become effective.
See also https://www.dkrz.de/up/services/analysis/data-processing/processing-on-mistral for more details.