Generate a run script
First step
There are several pre defined experiments available in the run-directory, listed as exp.<experiment-name>
.
To generate a run script for one of these experiments do
./make_runscripts <experiment-name> -r <script-directory>
Be careful not to forget script-directory and experiment-name. make_runscripts will generate all possible runscripts otherwise. make_runscripts only works after configuring the model, as it needs the proper information of the compiler used.
The resulting runscript exp.<experiment-name>.run
, for some experiments also a postprocessing script post.<experiment-name>.run
, will be stored in the run-directory and might need additional editing.
Use your experimenter ID
In general and to simplify collaboration it is an established way to copy a given script according to our personalized naming convention, see List of IDs.
Adaptations for accounting and parallelization
Example: Run the NextGEMS_R2B8 script as developer abc on Levante:
Copy the exp.NextGEMS_R2B8 experiment-template to your personal experiment-id:
cp run/exp.NextGEMS_R2B8 run/exp.abc0001 ./make_runscripts abc0001 -r run cd run
Edit the file exp.abc0001.run and change the model settings (account e.g. mh0287):
#SBATCH --account=mpiaes to #SBATCH --account=mh0287
#SBATCH --nodes=1 to #SBATCH --nodes=80
export OMP_NUM_THREADS=$((${SLURM_JOB_CPUS_PER_NODE%%\(*} / 4 / 1)) to export OMP_NUM_THREADS=4
Submit the job:
sbatch exp.abc0001.run
You can check the status of your jobs with:
squeue -u <user>
Results will be written to icon-mpim/experiments/abc0001
In case you want to run the full 10 days of this experiment you have to adapt the run time in the run script: Set the run time of the job to 3 hours
#SBATCH --time=03:00:00
— Monika Esch 2023/05/08 13:44