Graph partitioning

The mpas_tools.seaice.partition module contains a function for creating graph partition files for MPAS-Seaice that are better balanced than those created from Metis tools directly.

Running from compass

One way to run the tools is from compass using the files_for_e3sm test case.

This has the advantage that it can run with a version of ESMF that has been compiled with system compilers for compass. Compass also automatically downloads and links the files needed for determining the regions of sea-ice coverage. However, if you are unfamiliar with compass, there may be a learning curve involved in setting up and running the test case.

Conda environment

The other preferred way to use the sea ice partitioning tool is through the mpas_tools conda package. To install it, first install Mambaforge (if you don’t already have Miniconda3):

To activate it, run:

source ~/mambaforge/etc/profile.d/conda.sh
source ~/mambaforge/etc/profile.d/mamba.sh

To create a new conda environment for mpas_tools, run:

mamba activate
mamba create -y -n mpas_tools python=3.11 mpas_tools "esmf=*=nompi*"

This will create a new conda environment called mpas_tools that contains the mpas_tools package and also the version of ESMF without MPI support. This is necessary because the version with MPI support (the default) doesn’t typically work on HPC.

Each time you want to run the sea-ice partitioning tools, run:

source ~/mambaforge/etc/profile.d/conda.sh
source ~/mambaforge/etc/profile.d/mamba.sh
mamba activate mpas_tools

All the tools (including fix_regrid_output.exe built from the Fortran code) are part of the mpas_tools conda package.

You will also need an MPAS mesh file to partition. You do not need to pass a location for the MPAS cell culler (-c) or Metis (-g) because theses will be found automatically in the conda package

Graph partition tools

The tools prepare_seaice_partitions and create_seaice_partitions are used to create MPAS-Seaice graph partitions that are better balanced so that each processor “owns” cells from both polar and equatorial regions.

$ prepare_seaice_partitions --help
usage: prepare_seaice_partitions [-h] -i MESHFILENAMESRC -p FILENAMEDATA -m
                                 MESHFILENAMEDST -o OUTPUTDIR

Perform preparatory work for making seaice partitions.

options:
  -h, --help            show this help message and exit
  -i MESHFILENAMESRC, --inputmesh MESHFILENAMESRC
                        MPAS mesh file for source regridding mesh.
  -p FILENAMEDATA, --presence FILENAMEDATA
                        Input ice presence file for source mesh.
  -m MESHFILENAMEDST, --outputmesh MESHFILENAMEDST
                        MPAS mesh file for destination regridding mesh.
  -o OUTPUTDIR, --outputDir OUTPUTDIR
                        Output direct

The input mesh file contains the MPAS-Seaice mesh fields for the mesh used to create the “presence” file. The “presence” file itself contains an icePresence field that indicates where sea ice might be present. We typically use a 60-km mesh file and the corresponding presence file. The presence file will be regridded to the given output MPAS-Seaice mesh. Here the ice presence was determined as any cell in the mesh to have had ice present at any time during a 50 year MPAS-Seaice standalone simulation with the above 60-km mesh file. The output directory is often the current directory.

After this preprocessing has finished, the create_seaice_partitions tool can be run one or more times. It is significantly more efficient to provide a list of processor numbers than to call the tool for each processor number separately.

$ create_seaice_partitions --help
usage: create_seaice_partitions [-h] -m MESHFILENAME -o OUTPUTDIR
                                [-c MPASCULLERLOCATION] [-p OUTPUTPREFIX] [-x]
                                [-g METIS] [-n NPROCS] [-f NPROCSFILE]

Create sea-ice partitions.

options:
  -h, --help            show this help message and exit
  -m MESHFILENAME, --outputmesh MESHFILENAME
                        MPAS mesh file for destination regridding mesh.
  -o OUTPUTDIR, --outputDir OUTPUTDIR
                        Output directory for temporary files and partition
                        files.
  -c MPASCULLERLOCATION, --cullerDir MPASCULLERLOCATION
                        Location of MPAS MpasCellCuller.x executable.
  -p OUTPUTPREFIX, --prefix OUTPUTPREFIX
                        prefix for output partition filenames.
  -x, --plotting        create diagnostic plotting file of partitions
  -g METIS, --metis METIS
                        name of metis utility
  -n NPROCS, --nProcs NPROCS
                        number of processors to create partition for.
  -f NPROCSFILE, --nProcsFile NPROCSFILE
                        number of processors to create partition for.

The mesh filename provides the desired MPAS-Seaice mesh, the same as the destination mesh for prepare_seaice_partitions. The output directory is often the current directory. A directory containing the MpasCellCuller.x tool can be provided but by default it will be found in your path as part of the mpas_tools conda package. The output prefix will be prepended onto each graph partition file, and defaults to graph.info. The Metis tool is nearly always gpmetis, the default, and must be available in your path (which is the case if you use mpas_tools conda package). One graph partition file is created for each number of processors (one or more integers) provided. Alternatively, these can be listed, one value on each line, in a file. You can optionally save a NetCDF file with partition information partition_diag.nc, which will contain a partition_{nProcs} field for each number of processors requested.

Graph partition function

A helper function mpas_tools.seaice.partition.gen_seaice_mesh_partition() is used within create_seaice_partitions. It can also be called directly but must already have the files resulting from prepare_seaice_partitions available in the output directory.