Imperial HPC

Isca experiments can be submitted to Imperial's PBS scheduler by selecting the pbs scheduler:

from isca_tools.run import run_experiment

run_experiment(namelist_file, diag_table_file, scheduler='pbs')

The runner submits one PBS job per simulation chunk and makes each job depend on the previous chunk. Set n_nodes, n_cores, and max_walltime in the namelist's experiment_details section. max_walltime should be in HH:MM:SS format. The PBS request uses select=<n_nodes>:ncpus=<n_cores>:mpiprocs=<n_cores>. Add memory_per_node (for example, '16gb') to request memory per node; otherwise PBS uses its default of 1 GB per node.

Imperial uses routing queues, so the PBS scripts do not specify a queue. The Slurm-specific partition and nodelist options are not used. For current queue limits and resource advice, see Imperial's job sizing guidance.