...
For the dry core, on Harvard cluster, often writing to disk is the rate-limiting step. You can use -N 1 to ask slurm to put all your tasks on one node, and utilize the super fast, node-local disk /scratch. See example runscript: ~pchan/script/jetshift/fms-runscript.sbatch.
One can run an ensemble, by applying slightly different initial conditions using src/atmos_spectral/init/spectral_initialize_fields.f90. See example script: ~pchan/script/jetshift/ensemble.sh.