Many of the hpc machines have required local config changes, which we include in proteus/config/MACHINE.py. While the require changes are becoming less invasive, we are running into more users who have trouble submitting jobs and running tests on hpc machines. I think we could write into our machine configs both the MPI launch script (mpiexec, aprun, mpiexec_mpt, etc.) and a template job submission script (e.g. a PBS script template suitable for the machine). With that information, I think we could support something like this from the login node:
hpc_machine> parun myproblem.py
This would automatically submit an interactive batch job to the debug queue for 1 hour and run the given parun command with the required MPI launcher on 1 core (or n cores with a parun -n option)
parun myproblem.py -Q "-A ERDCV00898R40 -q standard -l walltime=001:00:00"
This would run the command in a batch job (or interactive job if -I is included).
Many of the hpc machines have required local config changes, which we include in proteus/config/MACHINE.py. While the require changes are becoming less invasive, we are running into more users who have trouble submitting jobs and running tests on hpc machines. I think we could write into our machine configs both the MPI launch script (mpiexec, aprun, mpiexec_mpt, etc.) and a template job submission script (e.g. a PBS script template suitable for the machine). With that information, I think we could support something like this from the login node:
hpc_machine> parun myproblem.pyThis would automatically submit an interactive batch job to the debug queue for 1 hour and run the given parun command with the required MPI launcher on 1 core (or n cores with a parun -n option)
parun myproblem.py -Q "-A ERDCV00898R40 -q standard -l walltime=001:00:00"This would run the command in a batch job (or interactive job if -I is included).