Sync meeting 2023 03 14 - multixscale/meetings GitHub Wiki

Sync meeting 2023-03-14

  • kickoff meeting (23-24 March)
    • what do we need to prepare?
      • Alan: lots
      • Overview of technical WPs (1+5+6) [Caspar, Kenneth]
        • 1 slide per WP task
        • focus on first year
        • work backwards from 1st year review
      • MultiXscale training and events for 2023 [Eli]
        • Introductory crash course on EESSI @ HPCKP'23? (17-18 May 2023, Barcelona)
          • Getting Access, Using EESSI, Structure/How it works + Motivation, Hands-On, apps (GROMACS), MPI (OSU), using GPUs (?), ...
          • 2h session?
        • sysadmin-focused tutorial on CVMFS (fully online)
        • something on using EESSI in CI pipeline?
          • towards scientific MXS WPs
          • could be part of course on "Best Practices"?
      • no voting for remote attendees
    • points to discuss
      • limited travel budgets vs demands from EU
        • EuroHPC Summit
        • CASTIEL2 kickoff meeting
  • Eli: asked about access to webpage
    • deliverable at end of March'23
    • templates for SC deliverables? via Alan
    • Alan can give access to website
    • basic homepage is in place at https://www.multixscale.eu
  • social media
    • Twitter account is there, no tweets yet https://twitter.com/multixscale
    • can figure out at kickoff how this will be handled
    • LinkedIn is TODO
    • can be managed by Susana (HPCNow!)
  • WP1 (SURF)
    • test suite: getting GROMACS test in shape, can serve as basis for other tests (OSU, TensorFlow)
      • could this be used for testing generic builds in EESSI
    • new EESSI pilot version
      • 2021.06 pilot version be removed (init script will fall back to 2021.12, with a warning)
      • new pilot version 2023.03 is WIP (still in eessi-hpc.org domain)
        • building compat layer basically works
        • would be nice to have strace in compat layer (does it need special permissions?)
          • can also be added to container
          • could help to figure out why using GPUs doesn't work in container
          • some problem with CUDA compat libraries
        • Gentoo Prefix bootstrap is working on x86_64 (+ ppc64le)
          • some problem for aarch64, but a fix is available
        • general consensus seems to be that compat layer should be stripped down to what's essential for software layer
  • WP5 (UGent)
    • bot
      • getting close to rewrite of interface between bot and compat/software layer
        • bot/build.sh script in compat/software layers that is used by bot (separation of concerns)
        • required PRs are open/being reviewed+tested, this setup already in use in NESSI projects
      • Thomas/Jonas/Kenneth/Bob are actively involvement in development
      • ~5 people are using bot in scope of NESSI project
      • required effort in MXS T5.3 (10PMs) is probably unestimated
      • maybe a bit more focus is needed on what's really needed in MXS T5.3
    • T5.1 support portal
  • WP6 (UB)
    • see above
  • WP7 (HPCNow!)
    • see above