...
Currently ony version 1.7.2 is installed, so before starting julia callJulia, load the corresponding module:
$ module load julia
Running Julia on the frontends
This is possible, but resources and runtime are limited. Be friendly to other users and work on the (shared) compute nodes!
Running Julia on the compute nodes
...
By default, MPI.jl will download and link against an own MPICH implementation. On the HLRN-IV systems, we advise using the Intel MPI implementation, as we have found some serious problems with the Open MPI implementation in conjunction with multithreading.
Therefore the julia Julia module set already some environment variables under the assumption that the impi/2018.5 module is used (both for MPI versions 0.19 or earlier and for the newer versions using the MPIPreferences system).
To add the MPI.jl package to your depot follow these step steps:
$ module load impi/2018.5
$ module load julia
$ julia -e 'using Pkg; Pkg.add("MPIPreferences")'
$; juliausing -e 'using PkgMPIPreferences; MPIPreferences.use_system_binary(); Pkg.add("MPI")'
You can test the the correct version is used via
$ julia -e 'using MPI; println(MPI.MPI_LIBRARY_VERSION_STRING)'
...
There is no direct dependency ti to impi/2018.5 in the juliaJulia's module file, so if needed it is still possible to adjust the environment to a different configuration before building and loading the MPI module.jl package. Please check to the MPI.jl documentation for details.
...
Also for the HDF5.jl package one of the HDF5 modules provided by the HLRN-IV system should be used. After loading an HDF5 module, copy the HDF5_ROOT environment variable to JULIA_HDF5_PATH:
$ export JULIA_HDF5_PATH=$HDF5_ROOT