AWS meeting 2023 03 09

EESSI/AWS sync meeting (2023-03-09)

• link to AWS project doc: https://docs.google.com/document/d/1CHG9fCh2LkfJ-EI8J-_Wr5NpHL5iwm8Wu6syfK9h7-c

Notes previous meeting

• 11 Jan 2023: https://github.com/EESSI/meetings/wiki/AWS-meeting-2023-01-11

Notes meeting 9 March 2023

• hands-on EESSI demo

  • on Ubuntu 22.04 VM in EC2 (c6g.2xlarge instance)
  • following updated "Getting Access" documentation (https://eessi.github.io/docs/getting_access/native_installation)
  • running EESSI demo scripts (https://eessi.github.io/docs/using_eessi/eessi_demos)

• sponsored credits

  • current batch ($10k) expires end of 2023Q1
    • we will have used ~45% of it
    • We need a fresh batch of sponsored credits for 2023Q2 & beyond
  • current burn rate is ~$1.5k/month
    • used for:
      • one Stratum-1 mirror server (EBS+EFS)
      • monitoring server (http://status.eessi-infra.org)
      • sources.easybuild.io (EasyBuild sources mirror)
      • testing of build-and-deploy bot being developed (CitC cluster)
      • build node(s) for next EESSI compat layer
  • Brendan => current batch of $10k credits has been extended until 2023-09-30
    • we should shout (in time) if more credits are required

• follow-up on discussion points of last meeting

  • S3-backed Stratum-1 mirror servers
    • should work, cfr. https://cvmfs.readthedocs.io/en/stable/cpt-repo.html#sct-s3storagesetup
    • but: client access still requires CernVM-FS, since data in S3 needs to be presented as (read-only) POSIX filesystem by CernVM-FS
    • can also consider using S3 for central Stratum-0 server
    • could look into AWS-specific Stratum-1 mirror servers (backed by S3) to improve user experience in AWS (lower latency)
  • ISC'23
    • EasyBuild tutorial proposal was not accepted
    • no Birds-of-a-Feather session on EESSI submitted (time constraints, unsure who will attend ISC'23)
    • Any opportunities for a booth talk on EESSI?
      • no AWS booth at ISC'23
      • Who involved with EESSI project will be attending ISC'23?
  • How to measure success
    • We have an idea on how to get some usage stats for EESSI (+ info on context in which its being used)
    • EESSI init script could send information to a "counting server"
    • incl. EESSI version, host info (OS, CPU), context (CI like GitHub Actions or not), anonymised user/host name combo, resolve IP to country, ...
    • there are some privacy concerns here...
    • discussion on whether this should be opt-in vs opt-out
    • not implemented yet, currently getting feedback
  • interest in HPC Tech Shorts episode on EESSI @ AWS?
    • maybe on the floor at ISC'23

• EESSI progress in last months

  • see also slides of latest EESSI update meeting: https://raw.githubusercontent.com/EESSI/meetings/main/meetings/EESSI_meeting_20230302.pdf
  • improved user-facing documentation
  • working towards a new EESSI pilot version (2023.03)
    • new compat layer (Gentoo Prefix)
    • rebuild of all software we've included in 2021.12 pilot version
    • Only use build-and-deploy bot to add software (no more manual building + ingesting)
    • Significantly expand software stack
      • More software
      • More recent compilers toolchains
      • NVIDIA GPU support
        • more info in https://eessi.github.io/docs/meetings/2022-09-amsterdam/EESSI-community-meeting-2022.09-014-GPU-support.pdf
  • development of build-and-deploy bot
    • see https://github.com/EESSI/eessi-bot-software-layer

• integration of EESSI in Parallel Cluster

  • would be wise to wait a couple of months with this

• help with EFA CUDA-aware support

  • wait until OpenMPI v5.0 is there
  • could be part of an EasyBuild Tech Talk on OpenMPI v5.0