SIG/HPC Meeting 2024-03-07¶

Attendees¶

Intel Driver -
Sherif is working on this, has a prototype, needs DKMS
- Used make spec script in the branch to create spec, and import from there
- We think that upstream should adopt a different format/packaging methodology
- Perhaps packit could be helpful?
What branch/version to use?
- rhel-specific branches say not to use them; use the 'backports' branches instead
- sherif appears to be in the right place
Next steps:
- Neil to bring dkms from epel into projects
- Sherif to upload to public location for review and testing
- Jeremy to work on testing with some latest hardware
AI SIG
- where will userspace tools live? HPC? AI? Both?
- Neil: it should be reasonable for us to have the ability to easily release a package in multiple SIGs
NVidia GPU driver Testing -
Did not get time to review Chris's work - will try to review this cycle
Kernel Cnode / MoS
re-actioning - Jeremy to work on once he has some time

Testing Warewulf - Brian
Current plan: put the tests upstream into Warewulf repo, Testing team can pull from / engage with upstream
- What precisely are we going to test?
- Functional/E2E tests -- provision a small cluster, etc (see last week's discussions)
- Future work can include e.g. slurm
- Chris to check on status of slurm
Packages to bring in
List on the wiki; needs updating (along with the rest of the wiki)
if anyone wants to bring something in, has questions, etc. Please ask/get in touch!
Neil to update the wiki

Vulnerability in lustre - related to user namespaces
Sherif was working on lustre-server, but it's a beast
DDN already builds RPMS, but... is it worth it to rebuild vs just use upstream?
- Sherif: thinks it makes sense to rebuild against our specific user/kernel space
- there are lustre-server for 8, but not 9, it appears.. why?
- documentation supports this but again.. why?
- Sherif to look into why lustre-server exists for 8 but not 9
Next meeting in two weeks on Thursday, March 1