Experiment details

The experiment_details namelist contains information on how to run the simulation, most of which is relevant for submitting jobs to a batch scheduler. It is an addition I made to the namelists in the Isca source code, and all required options must be specified to run the simulation using isca_tools.

Options

name

string
Name of experiment e.g. data saved in folder $GFDL_DATA/{name}.
You can use name='exp/run1' to set create a new exp directory and then save the run1 data within it. This may be useful for running similar experiments with different parameters.
Default: N/A

input_dir

string
Directory containing any input files required for the experiment such as namelist.nml, diag_table or co2.nc.
Default: N/A

n_months_total

int
Total duration of simulation in months.
Default: N/A

n_months_job

int
Approximate duration of each job of the simulation in months.
E.g. if n_months_total=12 and n_months_job=6, the experiment would be split up into 2 jobs each of length 6 months.
Default: N/A

n_nodes

int
Number of nodes to run the job on.
Default: N/A

n_cores

int
Number of cores for each node to run the job on.
Must be a power of \(2\) so typically \(8\), \(16\) or \(32\).
Default: N/A

memory_per_node

string
Optional memory request per node for PBS jobs, e.g. '16gb'. If omitted, PBS uses its default.
Default: N/A

resolution

string
Horizontal resolution of experiment.
Options are T21, T42 or T85.
Default: N/A

partition

string
Slurm queue to submit job to. Not used with PBS on Imperial HPC.
Default: N/A

overwrite_data

bool
If this is True and data already exists in $GFDL_DATA/{name}, then it will be overwritten.
If it is False and the data exists, an error will occur.
If a previous run failed part way through, running the same experiment again with overwrite_data=.false. will start using the last restart file created in that previous run. Typically set to False.
Default: N/A

compile

bool
If True, it will recompile the codebase before running the experiment.
Typically set to False.
Set this to True if you have made any changes to the underlying source code of Isca.
Default: N/A

max_walltime

string
Maximum time that the job can run. PBS on Imperial requires HH:MM:SS format, e.g. '01:00:00' for one hour. Slurm accepts additional formats.
Default: N/A

delete_restart_files

bool
If True, only the restart file for the final month will be kept. Otherwise, a restart file will be generated for every month.
Default: N/A

nodelist

string
Specify node to submit to on Slurm.
Options on debug partition on St Andrews kennedy are kennedy20, kennedy21 or kennedy22.
Number listed needs to match n_nodes.
kennedy[20-22] would request the 3 nodes 20, 21, 22.
If not given, will just submit to default nodes. If multiple jobs, this will likely result in jobs being submit in the wrong order, and an error occuring.
Default: N/A

This option is ignored when using PBS on Imperial HPC; PBS selects the nodes automatically.