# EESSI paper meeting (20210511) Call for papers: https://onlinelibrary.wiley.com/pb-assets/assets/page/journal/1097024x/SPE-SI-HPC-1607014410373.pdf Submission deadline: Monday **May 31st 2021** Overleaf project (contact Alan to get access): https://www.overleaf.com/project/60993d67cfe5ee67c2fe2fe0 ## Attendees - Alan O'Cais (Jülich Supercomputing Centre) - Kenneth Hoste (HPC-UGent) - Thomas Röblitz (HPC-UBergen) - Victor Holanda Rusu (CSCS) - Bob Dröge (HPC-UGroningen) - to invite - Terje - Caspar - Adam (review?) ## Follow-up sync meetings - Wed May 19th 13:30 - 15:00 CEST - Wed May 26th 13:30 - 15:00 CEST ## Attention points - Need to make clear what is different to what CC did. Check their paper. ## Places to steal from - Fenics proposals - NESSI / S4 - CZI grant proposal - FOSDEM submission - Alan's writeup - EESSI docs ## Structure - Introduction: Problem statement + motivation [Kenneth] - community aspect - changing HPC landscape - x86_64, aarch64, ppc64le, RISC-V - relevance to exascale? [Alan] - workflows - Project overview [Bob, Kenneth, Thomas] - design choices - overall structure & layers [Bob](Creating note...) - software projects (CernVM-FS, Gentoo Prefix, EasyBuild/Lmod/archspec & co) [Kenneth] - minimal requirements for clients - build containers (Singularity) - "continuous" release of software layer - bi-yearly for compat layer? - deployment with Ansible/Terraform [Bob, Terje?] - testing with ReFrame [Victor] - EESSI community [Kenneth] - How it started - Who's involved - Current structure - Lack of dedicated funding (bottom-up) - open source licensed, GitHub - Use cases [Thomas] - HPC cluster - Cloud - Workstation [Victor] - WSL, macOS - Testing across different systems - Building software on top [Alan] - Training (cluster in the cloud, magic castle) [Alan] - might take some use cases from S4 proposal - Work together with software developers - review installations - how to assess quality of installation (correctness, performance) - Facilitate testing in CI environments [Alan] - mount EESSI in GitHub Actions - Pilot setup demo [Bob] - 4 steps to get access in a Linux VM - focused effort: limited software, x86_64 + aarch64 - Discussion [Alan] - different CPUs (see intro?) [Kenneth] - limitations of CernVM-FS [Bob?] - risk of central setup (S0) - read-only - local disk (cache) - internet access (default setup, workaround possible) [Alan] - OS jitter - deployment workflow [Kenneth] - delay in adding stuff - workaround: build on top locally (temporarily) - security aspects [Kenneth] - separate section? - limit access to S0 - signed tarballs - automation - transparency - layers of security - proprietary software (EULAs, build vs run) [Kenneth] - Intel, CUDA, AOCC, ... - vs containers [Kenneth] - integration with vendor libraries (cfr. Cray, MPI) [Alan, Victor?] - compatibility of host & EESSI MPI library - OS jitter due to CernVM-FS + workaround (NFS export of CVMFS filesystem) [Victor] - OpenFOAM as example? - impact on other project in terms of improvements, adoption, etc. - contributions to EasyBuild/ReFrame [Victor] - adoption of CernVM-FS/Gentoo Prefix in HPC [Bob] - Evaluation [Alan] - benchmarks [Alan] - GROMACS (cfr. write-up by Alan) - scalable workload (OpenFOAM?) [Kenneth] - direct MPI eval: OSU (latency/bandwidth) - Acknowledgements [Thomas] - Fenics - Dell - ComputeCanada - NESSI + Terje (contributions, Stratum-1) - UGroningen (effort + Stratum-0 + S1) - SURF - HPC-UGent - JSC (testing) - AWS - Azure - OSU OSL - Future work [Kenneth] - automation (CI/CD) [Bob, Kenneth, Terje] - inject host libraries via /opt/eessi/lib + CVMFS variant symlinks [Alan] - MPICH toolchain [review Victor] - funding [Thomas] - flexibility in filesystem layer (CernVM-FS vs S3) + software layer (EasyBuild vs Spack vs ....) + also compat layer (cfr. Nix -> Gentoo move by CC) [Kenneth] - proper consortium [Thomas] - macOS [Kenneth, Terje]