Parallel execution

PyPARIS supports multiprocessing on one machine and MPI across processes that may run on multiple machines. Choose process counts to match your available resources and simulation configuration. The single-bunch tutorial gives the corresponding launch commands.

Local multiprocessing

The PyPARIS.multiprocexec launcher uses Python’s multiprocessing support and does not require an MPI installation. Its process count includes the master.

MPI environments

MPI execution requires both an MPI runtime and the Python mpi4py package. For a local Conda environment, one option is:

conda activate ecloud
conda install -c conda-forge mpi4py mpich

On a managed cluster, follow the site’s instructions for loading its MPI module and installing a matching mpi4py. Use the launcher belonging to that MPI installation; mixing a Conda MPI runtime and a cluster launcher can cause incompatibilities. Check what your active environment selects:

which mpiexec
python -c "from mpi4py import MPI; print(MPI.Get_library_version())"

The scheduler may require srun or another launch command instead of mpiexec. PyPARIS MPI execution does not require parallel HDF5.

See the mpi4py installation guide for supported installation methods.