How to get the init.nc file of 10-km mesh for MPAS 8.0

[wdiazcarballo]

We are benchmarking the 10-km mesh, provided here MPAS-Atmosphere mesh downloads, on MPAS-A model in the APAC HPC AI 2023 competition. We found some interesting execution profiles of MPAS 7.3 when analysing the profiles collected by the performance tool from the Barcelona Supercomputing Center. We were not sure whether some performance issues had already been addressed by the upgraded MPAS version and are willing to explore them.

We have successfully built the MPAS 8.0 init_atmosphere core. We have tried setting up the namelist_init for the 10-km unified mesh benchmark for MPAS 8 by reading the init.nc file of the MPAS7 benchmark. Due to the limited allowance SU and disk space we have for the competition, we have not successfully run init_atmosphere yet. There were issues with finding the input data such as CFSR and SST files (as the photo attached). So, I am just wondering if you might suggest where we can get the 10-km mesh with init.nc of MPAS 8. So that we can collect its profile and compare against the MPAS 7.3. Thank you!

 

[mgduda]

I think the MPAS v7.0 benchmark inputs from Index of /projects/mpas/benchmark/v7.0 should work with MPAS-A v8.0; but note that the look-up tables needed by the Noah LSM have changed between MPAS v7.x and MPAS v8.0, so you'll need to copy those into the unpacked benchmark case directory with something like:

cp ${MPAS_DIR}/src/core_atmosphere/physics/physics_wrf/files/* .

where ${MPAS_DIR} is set to the path of your compiled MPAS v8.0 code.

Alternatively, if you'd like to create your own initial conditions file using the same reanalysis as was used for the benchmark cases, you can download the CFSR:2010-10-23_00 file from https://www2.mmm.ucar.edu/projects/mpas/real/cfsr.2010102300.tar.gz .

 

[wdiazcarballo]

Thanks so much. We have copied the files to the 10-km unpacked benchmark folder as you suggested and successfully ran the atmosphere_model of MPAS 8.0.1 as the attached output file. From our experiments, the integration time of the MPAS 7.3 model is slightly better than that of the MPAS 8.0.1 model. Your helps allow us to be able to collect the execution profile which we will use to identify the factors that impacted execution performance on both versions of MPAS Atmosphere models.

 

[mgduda]

We'd very much appreciate hearing about any insights you gain regarding the performance reduction you're seeing in MPAS v8.0.1 compared with v7.3!

 

[jpiers]

I think the MPAS v7.0 benchmark inputs from Index of /projects/mpas/benchmark/v7.0 should work with MPAS-A v8.0; but note that the look-up tables needed by the Noah LSM have changed between MPAS v7.x and MPAS v8.0, so you'll need to copy those into the unpacked benchmark case directory with something like:

cp ${MPAS_DIR}/src/core_atmosphere/physics/physics_wrf/files/* .

where ${MPAS_DIR} is set to the path of your compiled MPAS v8.0 code.

Alternatively, if you'd like to create your own initial conditions file using the same reanalysis as was used for the benchmark cases, you can download the CFSR:2010-10-23_00 file from https://www2.mmm.ucar.edu/projects/mpas/real/cfsr.2010102300.tar.gz .

Perhaps somewhat related... I've noticed that Noah LSM option in namelist.atmosphere has changed to 'sf_noah' in MPASv8, from simply 'noah' in MPASv7. The latest User's Guide does not reflect this (https://www2.mmm.ucar.edu/projects/mpas/mpas_atmosphere_users_guide_8.0.1.pdf)

config_lsm_scheme = 'sf_noah'