meeting Mar 4 2021

• date & time: Thu Mar 4th 2021 - 2pm CET (1pm UTC)
  • (every first Thursday of the month)
• venue: (online, see mail for meeting link, or ask in Slack)
• agenda:
  • Quick introduction by new people
  • EESSI-related meetings in last month
  • S4 NeIC project proposal
  • Experiences with 2020.12 pilot repository
  • 2021.02 version of pilot repository: status
  • Updates on sponsorship by Azure/AWS
  • Call to action for 2021.02 pilot
  • Past & upcoming events
  • Q&A

Slides

• see https://github.com/EESSI/meetings/blob/main/meetings/EESSI_meeting_20210304.pdf

Meeting notes

(by Bob, Kenneth)

Introduction by new people

• New people on the call:
  • Kurt Lust (UAntwerpen + LUMI User Support Team)
  • Dennis Terhorst (Jülich Supercomputing Centre)
  • Alfred Gil (HPCNow!)
  • Pierre-Yves Defosse (LuxProvide)
  • Jörg Sassmannshausen (NIHR Biomedical Research Centre, UK)
• Kurt Lust has been working on a module file to set up the EESSI environment, see comment in https://github.com/EESSI/software-layer/issues/68
  • currently rather static, but good base to work on

S4 NeIC project proposal (effort led by Thomas Röblitz)

• More dedicated funding for EESSI
• Deadline: next Wednesday (Mar 10 2021)
• Decision: probably in August
• Alan will take a look at the project proposal (~10 pages)

Experiences with 2020.12 pilot repository

There are several options for finding the CUDA (driver) libraries. We need to figure out what works best, and which libaries we need.

• hard symlinks (not good enough for libcuda.so.1, path is different for different hosts)
• variable symlink (CernVM-FS feature), may require client configuration if path diverges from default target
• custom path /opt/eessi/lib we inject in glibc, requires symlinks to be added in host

Progress update

• master branches were renamed to main in all EESSI repositories

Filesystem layer

• no more CVMFS config repo for EESSI

  • no real disadvantage as long as we stick to the eessi-hpc.org domain
  • additional repositories will be picked up (because we use a single master key)

• set CVMFS_CLIENT_PROFILE=single instead of CVMFS_HTTP_PROXY=DIRECT in your default.local when using our new cvmfs-config packages on a single machine without a proxy; if you do have a proxy, still set CVMFS_HTTP_PROXY manually.

• Volunteers for setting up a Stratum 1 in AWS: Jörg

Compatibility layer

• QEMU allows us to run checks via Github Actions for all architectures, see: https://github.com/EESSI/compatibility-layer/pull/85/checks?check_run_id=2007040272
• Open issue about a more extensive test suite for compatibility layer, which can be run from the playbook, Github Actions, cron jobs, etc: https://github.com/EESSI/compatibility-layer/issues/42

Software layer

• Kenneth did a first attempt to create a Github Action that can do quick test runs of some apps using the test scripts in the eessi-demo repo: https://github.com/EESSI/eessi-demo/actions/runs/611850589
• Similarly, the following Action uses easystacks to check the (installation) status of the software layer, by checking for missing modules: https://github.com/EESSI/software-layer/pull/74

Infrastructure repo

• Scripts/tools to deploy both dynamic (e.g. build nodes) and static (e.g. login node, Stratum 1) virtual machines
• It would be useful to organize a training session at some point to demonstrate how all of this works
• For now, Terje keeps playing with this, probably not worth using PR workflow just yet (just push to main branch)

2021.02 version of pilot repository

• There are compat layers for each CPU arch, but they are some unresolved issues (upstream Lmod package needs to be unmasked for aarch64 + ppc64le, configuring glibc to pick up on /opt/eessi/lib isn't work yet).
• No installations in software layer yet (but necessary preparations were made)
• We are probably going to wipe this installation and redo it as version 2021.03.
• For the software layer we are probably going to support the same hardware targets and more or less the same software.
• Kenneth is figuring out with Nvidia if we can redistribute CUDA in our repo, and if not, what the best workaround is...

Updates on sponsorship by Azure/AWS

• We are making good use of the AWS credits
• ~$600 worth of credits spent in AWS in Feb'21 (plenty more sponsored credits left in AWS)
• No updates on the Azure sponsorship...

Call to action for 2021.02 pilot

• If you want to help, let us know on Slack!
• Setting up additional Stratum 1 servers + proxies in AWS
• Building software for different CPU targets in 2021.03 (probably) version of pilot repo

Past & upcoming events

• If anyone has EESSI meetings/presentations planned or knows of interesting opportunities, let it know!

Q&A

• Jörg: can we add more bioinformatics tools?
  • We deliberately started with a very minimal list that just contains a few applications from different fields. But, especially if it attracts more people, we can add a few more tools. If you want this, try to test it locally and open a PR to have it included in the build script. We are still working on better documentation for the build process.