Version 10 (modified by jgipsl, 9 years ago) (diff) |
---|
How to run a simple test case with ORCHIDEE offline with a batch job
First prepare a run directory following the how to How to run a simple test case with ORCHIDEE.
Then prepare a job according to computer environement and the choice of parallelization and submit it.
curie/TGCC
MPI mode
- First prepare a run directory following the how to How to run a simple test case with ORCHIDEE. But instead of using wget to copy the files, they can be found in the directory /ccc/work/cont003/dsm/p86ipsl/IGCM and can be linked or copied directly. The path is the same.
- Then create a text file as follow and save it as Job_curie :
#!/bin/ksh ###################### ## CURIE TGCC/CEA ## ###################### #MSUB -r test # name of the job #MSUB -o Script_Output # name of output file for standard messages #MSUB -e Script_Output # name of output file for error messages #MSUB -eo #MSUB -n 4 # Request numbre of cores #MSUB -T 1800 # Time limit in seconds #MSUB -Q test # Queue test, priority acces #MSUB -q standard #MSUB -A gen6328 # Set your project id BATCH_NUM_PROC_TOT=$BRIDGE_MSUB_NPROC set +x cd ${BRIDGE_MSUB_PWD} /usr/bin/time ccc_mprun -n 4 ./orchidee_ol
- Submit the job :
ccc_msub Job_curie
- Survey the job while it is still in queue :
ccc_mstat -u
- Kill the job if needed, take the jobid from the ccc_mstat command :
ccc_mdel jobid
Hybbrid MPI/OpenMP mode
Note that the hybrid mode is only for the coupled mode with lmdz.
#!/bin/bash #MSUB -r test # Request name #MSUB -n 32 # Total number of tasks to use (MPI) #MSUB -c 2 # Number of threads per task to use (OMP) #MSUB -N 4 # Number of nodes = 32*2/16 = 4 #MSUB -T 1800 # Elapsed time limit in seconds #MSUB -q standard # Choosing thin nodes #MSUB -Q test # Choosing test Queue #MSUB -o Script_output # Standard output #MSUB -e Script_output # Error output #MSUB -A gen**** # submission group number set -x cd ${BRIDGE_MSUB_PWD} export KMP_STACKSIZE=2g export KMP_LIBRARY=turnaround export MKL_SERIAL=YES export OMP_NUM_THREADS=2 # = number of thread ask in header (2 - 4 - 8 ) ccc_mprun ./lmdz.x
ada/IDRIS MPI mode
#!/bin/ksh # ###################### # ## ADA IDRIS ## # ###################### # Job name # @ job_name = SECHSTOM # Job type # @ job_type = parallel # Standard output file name # @ output = Script_Output_SECHSTOM.000001 # Error output file name # @ error = Script_Output_SECHSTOM.000001 # Total number of tasks # @ total_tasks = 4 # @ environment = "BATCH_NUM_PROC_TOT=4" # Maximum CPU time per task hh:mm:ss # @ wall_clock_limit = 1:00:00 # End of the header options # @ queue # Temporary due to memory problems on mpi version export MP_EUILIBPATH=/smplocal/lib/ibmhpc/pe12012/ppe.pami/gnu/lib64/pami64 date /usr/bin/time poe ./orchidee_ol date
- Submit the job :
llsubmit Job
- Survey the job while it is still in queue :
llq | grep login
- Kill the job if needed, take the jobid from the llq command :
llcancel jobid
obelix/LSCE MPI mode
###################### ## OBELIX LSCE ## ###################### #PBS -N MyTest #PBS -m a #PBS -j oe #PBS -q medium #PBS -o Script_Output #PBS -S /bin/ksh #PBS -v BATCH_NUM_PROC_TOT=4 #PBS -l nodes=1:ppn=4 ## Load same netcdf as used in the compilation . /usr/share/Modules/init/ksh module unload netcdf module load netcdf/4p ## Go to current directory cd $PBS_O_WORKDIR ## Launch the model mpirun -np ${BATCH_NUM_PROC_TOT} orchidee_ol
- Submit the job :
qsub Job
- Survey the job while it is still in queue :
qstat -u login
- Kill the job if needed, take the jobid from the qstat command :
qdel jobid