Meeting with BioCommons (2023 03 10)

2023-03-10 EESSI/Australian/BioCommons catch-up

(see also https://docs.google.com/document/d/1RGRiGbmzeJ_RfBjqxmpaFOZDuSRFlSYyW_Z3F0nY_Hc)

Attendees

EESSI

• Kenneth Hoste (HPC-UGent)
• Alan O'Cais (CECAM)

BioCommons BYOD project

• Hoylen Sue(QCIF)
• Peter Marendy (QCIF)
• Steffen Bollmann (Univ. of Queensland)
  • NeuroDesk
• Alex Ip (AARNet)
• Steele Cooke (AARNet)
• Greg D’Arcy (AARNet)

Q&A

Steffen:

• Why was Gentoo chosen as a compatibility layer instead of for example Singularity/Apptainer?
  • Performance
  • Limiting what needs to be downloaded to run (thank you CVMFS)
  • Combining multiple apps
  • Access to specialized hardware like Infiniband, GPUs, ...
  • Security (ingesting binaries from others)
• Is the complex build process (and the difficult recipe definition) limiting the addition of tools?
  • Yes, see for example https://github.com/easybuilders/easybuild-easyblocks/pull/2248 (make CMake behave)
  • Controlling the builds also has benefits (cfr. how we handle CUDA)
• How is it ensured that the software does not contain malicious code that for example steals ssh-keys?
  • Same situation as containers, installing software for cluster on NFS filesystem, ...
  • See also https://github.com/EESSI/filesystem-layer/issues/131 (vulnerability scanners)
• Is there one catalog for everyone or can sites customize which software is offered?
  • Lmod can hide modules, see https://lmod.readthedocs.io/en/latest/140_deprecating_modules.html + isVisible hook (https://lmod.readthedocs.io/en/latest/170_hooks.html )
  • There will be good user-facing documentation to build stuff on top of what EESSI provides
• How are bugs in the build process handled that are discovered after people are already using a tool and then require rebuilding a tool with the same version number - is this tool replacing the existing faulty software or is there a mechanism to warn users?
  • Could print warning when module is loaded
• How is ensured that the software built is working correctly (e.g. running tests with test data and comparing outputs?)
  • Sanity check commands in EasyBuild
  • EESSI test suite, see https://eessi.github.io/docs/meetings/2022-09-amsterdam/EESSI-community-meeting-2022.09-020-software-testing.pdf + https://github.com/EESSI/test-suite
• Long-term funding of the work—what are the costs, what is the sustainability plan?
  • Short-term: MultiXscale EuroHPC project (2023-2026)
  • Longer-term: payed “subscription” to EESSI (better support, help with adding software, …)
  • Explore other funding models
  • Collaboration with consultancy companies like HPCNow!
  • Sponsored credits by AWS/Azure

Greg:

• EuroHPC project background & requirements
  • https://eessi.github.io/docs/meetings/2022-09-amsterdam/EESSI-community-meeting-2022.09-021-MultiXscale.pdf
• Funding model / Sustainability

Hoylen:

• What are the plans for teaching new and novice users how to use it (e.g. documentation, training)?
  • EESSI docs: https://eessi.github.io/docs/using_eessi/eessi_demos
  • Training is an explicit goal in MultiXscale
  • Model like Carpentries
  • EESSI tutorial at ISC/SC at some point
• How do users discover what is available in EESSI, and how is that maintained?
  • “module avail”
  • Software overview in EESSI documentation (for example via JSON data dumped by Lmod)

Alex:

• How do you control who adds/updates what in which repo?
• Is software patched regularly? (partially answered for compatibility layer in Steffen’s questions)
• Can Australia participate (like Eurovision)? (e.g. Australian Stratum 1)
  • Short answer: yes, EESSI is (worldwide) community project
  • Joint funded projects are worth exploring
  • Some opportunities to work together on compat layer (like EESSI does with Canadian HPC sites), CVMFS, ...

Steele:

• How do you implement storage on the S0/S1 e.g. do you use a local disk or a remote S3 bucket?

  • See also https://hackmd.io/7otdq5tSRuyIWAJaF7NH6A

• https://github.com/ComputeCanada/puppet-magic_castle