Getting started with Julia on a new cluster can sometimes be a challenge. Below we provide some common tips that will hopefully help you with your onboarding process.
When starting on a new HPC cluster the first thing you should do is figure out if there is a pre-configured Julia module (Lmod module) available on the cluster. To that end, module key julia or module spider julia might be helpful commands.
If there is no Julia module available that you can load, you should download and use the regular, precompiled Julia binaries. We strongly recommend to use juliaup for this. Alternatively, you can also manually download the binaries directly from the website. In any case, you should generally not build Julia from source (unless you have a very good reason).
Note that you do not need root privileges. Julia, and its packages, works great without special permissions in user space.
One you have Julia installed and you can run julia from the command line, you should place the Julia depot - the .julia folder where Julia stores all dependencies, logs, etc. - on an appropriate file system. By default, it will be stored in $HOME/.julia. This may or may not be a good choice, but more often than not it isn't.
You want to choose a file system with the following properties
no tight quotas (at least >= 20 GB)
read and write access (ideally also from compute nodes)
good (parallel) I/O
no automatic deletion of unused files (or otherwise you have to find a workaround)
On most clusters these criterion are best fit on a parallel file system (often $SCRATCH). In this case, you should put JULIA_DEPOT_PATH=$SCRATCH/.julia into your .bashrc (and your job scripts if .bashrc is not loaded by non-interactive jobs).
Note: If the last point (automatic deletion of unused files) is an issue for you, a pragmatic workaround could be a cronjob that touches all files in the Julia depot every once in a while.
JULIA_CPU_TARGET appropriately.On many clusters, the sections above are all you need to get a solid Julia setup. However, if you're on a heterogeneous HPC cluster, that is, if different nodes have different CPU (micro-)architectures, you should/need to do a few more preparations. Otherwise, you might encounter nasty error messages like "Illegal instruction".
To make Julia produce efficient code that works on different CPUs, you need to set JULIA_CPU_TARGET. For example, if you want Julia to compile all functions (clone_all) for Intel Skylake (skylake-avx512), AMD Zen 2 (znver2), and a generic fallback (generic), for safety, you could put the following into your .bashrc:
export JULIA_CPU_TARGET="generic;skylake-avx512,clone_all;znver2,clone_all".
You can get the CPU target name for the current system with Sys.CPU_NAME. For more information, see this section of the Julia documentation and this section of the developer documentation.