Azure meeting 2023 06 19

EESSI/Azure sync meeting 2023-06-19

attending:

• Alan, Davide, Martin, Kenneth, Laura

Agenda:

• Quick evaluation of ISC'23

  • talk by Elisabeth (HPCNow!) at Azure booth

• Update on OCRE project with Azure/Oracle (Ivar)

• Current status of EESSI

  • New pilot version 2023.06 is in the works
    • 2023.04 basically abondended because too much problems popped up with OpenSSL 3.x in compat layer
    • 2023.06 effort started
      • compat layer is in place w/ OpenSSL 1.1
      • first builds done, about to be deployed - see https://github.com/EESSI/software-layer/pull/260
      • software will be mix of what was in EESSI pilot 2021.12 + what's needed for MultiXscale (ESPResSo, walBerla, LAMMPS, OpenFOAM, ...) + beyond that
      • doing what's necessary to get more people involved with building out software layer along with build-and-deploy bot
  • Good progress on portable test suite: https://github.com/EESSI/test-suite
    • blueprint test with GROMACS as starting point for portable test suite using ReFrame
    • hoping to also use this to report on how good performance for a particular app is by collecting performance data across different systems

• builds with Intel compiler compiler?

  • EULA of oneAPI compiler suite may allow redistribution

• support portal for EESSI is being set up

  • based on GitLab Service Desk

• New colleagues at HPC-UGent + SURF for MultiXscale projects

• Looking into new Slurm cluster in Azure, to be used by build-and-deploy bot

  • Current CycleCloud instance is not exactly easy to manage (and specific to Azure)...
  • Maybe Magic Castle is easier?
    • Also "neutral" tool, can employ same tools in AWS (since Cluster-in-the-Cloud seems like a dead end)
  • azhop spins up more stuff, also uses CycleCloud internally
  • ParallelCluster is another option, but specific to AWS

• Interest in AMD Genoa X (zen4) instances as build nodes

  • general availability since Fri 16 June
  • min. recommended version for compiler is GCC 13.1 or AOCC 4.x
    • no plans yet for common foss version based on GCC 13.x - see https://github.com/easybuilders/easybuild-easyconfigs/pull/18066
    • foss/2023a will be based on GCC 12.3

• AWS HPC Tech Short on EESSI

  • https://www.youtube.com/watch?v=Fzv4ieiI1jo
  • Short and to the point (< 8min), captures essence of what EESSI is about quite well

• EasyBuild Tech Talk (https://easybuild.io/tech-talks)

  • Cloud HPC talk from UK (Stefano from University of Bath) about moving completely to the cloud
  • rehash of recent talk at HPC-SIG UK meeting, with more information about EasyBuild aspect
    • see https://hpc-sig.org.uk/index.php/event/hpc-sig-meeting-tuesday-13th-june-2023-university-of-bath-hybrid/
  • currently stuck in "session to be planned" phase...

• Alan: More GitHub runners

  • https://github.com/APN-Pucky/gentoo-prefix-tester

• next sync meeting

  • Mon 17 July 2023: Kenneth & Alan will be afk => cancelling that one
  • Mon 21 Aug 2023: OK for Kenneth, Martin, Alan

Notes of previous meetings

• https://github.com/EESSI/meetings/wiki/Azure-meeting-2023-05-15
• https://github.com/EESSI/meetings/wiki/Azure-meeting-2023-04-17
• https://github.com/EESSI/meetings/wiki/Azure-meeting-2023-02-17
• https://github.com/EESSI/meetings/wiki/Azure-meeting-2023-01-20
• https://github.com/EESSI/meetings/wiki/Azure-meeting-Dec-9-2022