123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475 |
- !
- ! Settings for SLURM job manager on NOAA theia - for details and other
- ! options: "man sbatch" and
- ! https://rdhpcs-common-docs.rdhpcs.noaa.gov/wikis/rdhpcs-common-docs/doku.php?id=slurm#migration_to_slurm_workload_manager
- !
- ! queue system (bsub,loadleveler)
- queue : slurm
- ! (1) passed directly (ie "as is" at the command line)
- !------------------------------------------------------------------
- ! (--share or -s for example to share node(s) with other jobs)
- ! add here also the long options
- queue.slurm.submit.options :
- !queue.slurm.submit.options : --share
- ! Note that this is added to the 'submit.options' key value from the pycasso-tm5-expert.rc
- ! (2) Short options set by end user (and processed by submit_tm5_tools.py script
- !----------------------------------------------
- ! First, list of queue options (same for each step):
- ! These each get prefixed with
- queue.slurm.options : job-name account time ntasks nodes output
- queue.slurm.option.output : %x.o%j
- ! job name:
- queue.slurm.option.job-name : ${job.name}
- ! NOAA-specific modification, appends jobstep number to job name; see
- ! TM5/base/branches/theia_slurm/bin/submit_tm5_tools__noaa_slurm.py
- queue.slurm.name.append.jobstep : T
- ! The "-A co2" option is specific to NOAA theia "co2" project
- queue.slurm.option.account : co2
- ! queue.slurm.option.qos : batch
- ! Partition should be "batch" for parallels jobs and "shared" for
- ! serial jobs. Note that bigmem is also a valid option. The partition
- ! will be automatically set based on ntasks, unless it is set in the
- ! RC file.
- !queue.slurm.option.partition : shared
- !queue.slurm.option.partition : batch
- ! NOTE: o and e options are not active because they are not in the
- ! options list. They are here left as placeholders, showing use of
- ! <auto>. This was handled by special code (now commented out) in
- ! TM5/base/branches/theia_slurm/bin/submit_tm5_tools__noaa_slurm.py.
- ! That was for some other previous slurm platform/implementation, we
- ! are going to try using the default output file name.
- ! queue.slurm.option.o : <auto>
- ! queue.slurm.option.e : <auto>
- ! JOB TASKS, NODES, RUNTIME, AND MEMORY REQUIREMENTS
- !
- ! Modify as appropriate for your particular run.
- ! For a serial job, put something like the following in your RC file.
- ! NOTE: no N (no. of nodes), and ntasks MUST be 1
- !queue.slurm.option.ntasks : 1
- !queue.slurm.option.time : 0:50:00
- !queue.slurm.option.mem : 5g
- ! For a parallel MPI job, put something like the following in your RC file.
- ! NOTE: lack of "mem" option; not used for exclusive-node scheme for parallel jobs.
- !queue.slurm.option.ntasks : 48 ! this is the number of cores
- !queue.slurm.option.nodes : 4 ! this is the number of nodes (would result in 12 cores used per node)
- !queue.slurm.option.time : 4:30:00
|