Skip to content

EESSI software stack

EESSI is the software stack of the EuroHPC Federation Platform (EFP) and documented in the EFP documentation with specific instructions for running on LUMI. That page also mentions specific restrictions for LUMI that will change over time.

EESSI is also a broader project with the software stack available on many non-EuroHPC systems. You can even run it on most Linux workstations, Windows systems with WSL2 and macOS systems with a Linux VM/container solution, but the machine needs to be connected to the internet when running the software. On LUMI however, we have to deploy EESSI in a special way (avoiding the CernVM-FS filesytem usually used for EESSI) due to the architecture of the LUMI machine and measures taken to limit OS jitter, so not everything in the regular EESSI documentation may apply to LUMI.

Support is only offered through the EFP helpdesk. LUST cannot forward these requests.

EESSI also has its own training events and a weekly EESSI happy hour session. You'll find materials from past training events there also.

The EESSI software stack is currently not offered through modules. Access is also different on the login nodes and on the compute nodes. For the expert: EESSI on LUMI is currently not offered through CernVM-FS but through SquashFS files that are synchronised with the main EESSI repository daily. On the login nodes, you can use EESSI through a container while on the compute nodes it is fuse-mounted when a job requests it.

The EFP software catalog requires logging in with an EFP account, but a second version of the EESSI software catalog is available in the EESSI documentation.

It is possible to install additional software on top of the EESSI stack. In fact, many EasyBuild recipes for the foss toolchain and its subtoolchains should work. These are regular EasyBuild recipes and should be supported by the EasyBuild community, not by LUST. LUST also does not plan to do any development on top of EESSI and will continue to work on its own software environment built on top of the HPE Cray Programming Environment for which we have full vendor support, and using Cray MPICH rather than Open MPI as Cray MPICH is guaranteed to work well on the LUMI interconnect and integrates properly with the scheduler.