E3SM
E3SM-IO is the parallel I/O kernel from the E3SM climate simulation model. It makes use of PIO library which is built on top of PnetCDF.
This benchmark currently has two cases from E3SM, namely F and G cases. The F case uses three unique data decomposition patterns shared by 388 2D and 3D variables (2 sharing Decomposition 1, 323 sharing Decomposition 2, and 63 sharing Decomposition 3). The G case uses 6 data decompositions shared by 52 variables (6 sharing Decomposition 1, 2 sharing Decomposition 2, 25 sharing Decomposition 3, 2 sharing Decomposition 4, 2 sharing Decomposition 5, and 4 sharing Decomposition 6).
You can find more information in Parallel-NetCDF GitHub repository.
Configuration
You can configure the ES3M-IO benchmark with the following options. Notice that if you use the configuration.json
approach to define the runs for h5bench
, we will automatically generate the final configuration file based on the options you provide in the JSON file. For standalone usage of this benchmark, you can refer to E3SM-IO repository.
Parameter |
Description |
---|---|
|
Keep the output files when program exits |
|
I/O strategy to write ( |
|
I/O library name to perform write operation ( |
|
Number of records/time steps for F case h1 file |
|
Enable write performance evaluation |
|
Define the HDF5 compression algorithm to use |
Warning
h5bench temporarily only supports -x blob
and -a hdf5
. If you set other options, they will be overwritten to the supported version.
JSON Configuration (recomended)
To run an instance of AMReX HDF5 benchmark you need to include the following in the benchmarks
property of your configuration.json
file:
{
"benchmark": "e3sm",
"file": "coisa.h5",
"configuration": {
"k": "",
"x": "blob",
"a": "hdf5",
"r": "25",
"o": "ON",
"netcdf": "../../e3sm/datasets/f_case_866x72_16p.nc"
}
}