Now that we’ve built a cluster, let’s install WRF:
We’re going to install WRF on the HeadNode, we’re able to do this as the architecture of the HeadNode instance type, c6a.2xlarge
, matches the compute nodes so Spack does the correct microarchitecture detection. In most other cases it makes sense to install on compute nodes.
spack install -j $(nproc) wrf@4.3.3%intel build_type=dm+sm ^intel-oneapi-mpi+external-libfabric
The command spack install -j $(nproc) wrf%intel build_type=dm+sm ^intel-oneapi-mpi+external-libfabric
tells Spack to install WRF using the latest version in the Spack recipe. It passes some build flags:
Spack Flag | Description |
---|---|
-j $(nproc) |
Compile with all the cores on the HeadNode. |
@4.3.3 |
Specify version 4.3.3 of WRF. |
%intel |
Specify the Intel Compiler (icc) we installed in e. Install Intel Compilers. |
build_type=dm+sm |
Enable hybrid parallelism MPI + OpenMP. |
^intel-oneapi-mpi+external-libfabric |
Uses Intel MPI which we added in e. Install Intel MPI. |
This will take about 3 minutes to install. While that’s installing feel free to advance to the next step and pull down the Conus 12-km model.