-
Notifications
You must be signed in to change notification settings - Fork 858
WeeklyTelcon_20200526
Geoffrey Paulsen edited this page May 26, 2020
·
1 revision
- Dialup Info: (Do not post to public mailing list or public wiki)
- Jeff Squyres (Cisco)
- Austen Lauria (IBM)
- Aurelien Bouteiller (UTK)
- Barrett, Brian (AWS)
- Brendan Cunningham (Intel)
- Christoph Niethammer (HLRS)
- Edgar Gabriel (UH)
- George Bosilca (UTK)
- Harumi Kuno (HPE)
- Howard Pritchard (LANL)
- Joseph Schuchart
- Joshua Ladd (nVidia/Mellanox)
- Matthew Dosanjh (Sandia)
- Michael Heinz (Intel)
- Naughton III, Thomas (ORNL)
- Ralph Castain (Intel)
- Todd Kordenbrock (Sandia)
- Geoffrey Paulsen (IBM)
- David Bernhold (ORNL)
- Josh Hursey (IBM)
- William Zhang (AWS)
- Akshay Venkatesh (NVIDIA)
- Artem Polyakov (nVidia/Mellanox)
- Brandon Yates (Intel)
- Charles Shereda (LLNL)
- Erik Zeiske
- Geoffroy Vallee (ARM)
- Mark Allen (IBM)
- Matias Cabral (Intel)
- Nathan Hjelm (Google)
- Noah Evans (Sandia)
- Scott Breyer (Sandia?)
- Shintaro iwasaki
- Xin Zhao (nVidia/Mellanox)
- mohan (AWS)
- Building with older libfabric fails, which causes all Coverity builds to also fail.
- libfabric is preinstalled, should be an easy fix on aws.open-mpi.org
- new selection code put in by AWS broke older libfabrics.
- should continue to support back to v1.0.5
- AWS will fix
- Jeff filed 7753 master-only.
Blockers All Open Blockers
Review v4.0.x Milestones v4.0.4
- v4.0.4rc2 - tags, and push up to github.
- Couldn't run the tarball builder. Something's broken with it in Jenkins
- Configure in jenkins there's a bad URL to get the
- It's now stuck. Best case brian can look at tallbar builder Friday.
- Jenkins Problem is the script Jenkins kicks off ompi-scripts repos.
- Jeff, Howard, and Brian have access.
- Brian played with it a bit on Saturday.
Review v5.0.0 Milestones v5.0.0
-
PR7762 - Giles filed - we're missing some fortran and C functions for MPI_Status
- Jeff is reviewing.
- Possibly wanted on v4.0.x
-
Schedule:
- PRRTE - Intel is changing his priority to move PRRTE to low priority.
- Intel needs it by Q4, but end of summer at the latest.
- It works with tiny option set.
- Many feature requests. Many CI options.
- What are all the fetures? - many ci tests
- Branching early, life just gets more painful.
- One silver lining - is most of these features are PRRTE alone.
- So submodule updates are pretty easy. Doesn't require cherry-picks.
- If there's a feature that's blocking functionality
- Some things would be considered as major regressions against OMPI v4
- May need an audit - what features aren't there
- What's no there with prrte launch?
- judge priority of missing features, open issues, etc.
- SOMEONE could do this, but wouldn't neccisarily give a good indication of v4-vs-v5 features.
- May need an audit - what features aren't there
- If we drop OMPI v5.0 without a number of v4 features.
- HUGE outcry from community if we don't support a lot of features from v4.
- We are dying under the weight of our feature set.
- Branching is not releasing. But putting in large features is painful after branch.
- Everyone is approaching from "my company's" point of view, but we're managing the software for the community.
- Both points of view are correct.
- Set of things that none of our companies need, but that the community needs.
- Bigger picture - how do we maintain this code? Haven't been able to maintain.
- Right now we're saying "we need to do a release, lets see what we've got, and add more later".
- Bigger picture - we could put together a README.md explaining why we removed some features.
- Help set community's expectations.
- Scope vs Schedule vs Resources
- Probably can't
- Can't release v5.0 because need to have these discussion.
- Can't release v5.0 without new runtime.
- Only other possible solution might be a v4.1 branched off of v4.0.x somewhere
- Would need VERY specific cherry-picking into, and avoid any runtime issues.
- One of the v5.0 RMs will have a list by Monday of needed features.
- Moved the branch date to June 2nd, but will discuss next week with more data.
- Original release was planned for end of June. Will discuss more next week
- PRRTE - Intel is changing his priority to move PRRTE to low priority.
- Intel needs it by Q4, but end of summer at the latest.
- It works with tiny option set.
- Many feature requests. Many CI options.
- MTT on master is looking pretty good.
- A recent commit broke some CI using older libevent.
- AWS is working to fix this.
- scale-testing, PRs have to opt-into it.