Conference call notes 20190904

(back to Conference calls)

Notes on the 131th EasyBuild conference call, Wednesday Sep 4th 2019 (17:00 - 18:00 CEST)

Attendees

Alphabetical list of attendees (5):

• Hugo Meilan (Microsoft HPC)
• Alan O'Cais (Jülich Supercomputing Centre)
• Mikael Öhman (Chalmers University of Technology, Sweden)
• Bart Oldeman (ComputeCanada)
• Davide Vanzo (Vanderbilt University, USA)

Agenda

• Update on progress towards EasyBuild 4.0
• update on 2019b common toolchains
• Q&A

Progress towards EasyBuild 4.0

• ETA:
  • Slightly delayed wrt initial expectations. The main roadblock is the OpenMPI easyblock changes that are preventing the release of 2019b toolchains with 4.0.0.
• Current status:
  • Fully replaced setuptools.setup with distutils.core.setup for reliability
  • Fixed version check compatibility with Python 3
  • Fixed --from-pr using archived easyconfig files
  • Fixed Python 3 compatibility issues for easyblocks using .items() and sys.maxint methods
• Mikael: The current separation between icc and ifort creates issues when using HMNS since both modules must be loaded in order for some software to work correctly. It would be nice to move the MODULEPATH change to iccifort instead. Some work has been done on this already, but there are some issues building intelcuda on top of it.
  • Davide: Since this is a fundamental change, we might be too tight to properly test it and release it with 4.0.0.

2019b common toolchains

• The only blocker is the OpenMPI easyblock that prevents foss/2019b to be merged. Everything else seems fine. Maintainers should definitely focus on testing and merging the corresponding PR.

Q&A

• Davide: Hugo, can you provide some context for your interest in EasyBuild?
  • Hugo: We are trying to help customers by providing a packages software infrastructure on our HPC cluster.
  • Currently we are focusing on building and optimize toolchains for both Intel and AMD architectures. Then we will build optimized applications.
  • Managed to build RPM packages with EasyBuild without issues.
  • Still have to define how to provision architecture specific stacks on compute nodes.
    • Mikael: CVMFS could be a good option to consider.