-
Notifications
You must be signed in to change notification settings - Fork 859
WeeklyTelcon_20210831
Geoffrey Paulsen edited this page Aug 31, 2021
·
1 revision
- Austen Lauria (IBM)
- Brendan Cunningham (Cornelis Networks)
- Geoffrey Paulsen (IBM)
- George Bosilca (UTK)
- Harumi Kuno (HPE)
- Hessam Mirsadeghi (NVIDIA))
- Jeff Squyres (Cisco)
- Joseph Schuchart (HLRS)
- Josh Hursey (IBM)
- Marisa Roman (Cornelius)
- Matthew Dosanjh (Sandia)
- Michael Heinz (Cornelis Networks)
- Sam Gutierrez (LANL)
- Siripaul (Intel)
- Thomas Naughton (ORNL)
- Todd Kordenbrock (Sandia)
- Tomislav Janjusic (NVIDIA)
- William Zhang (AWS)
- Akshay Venkatesh (NVIDIA)
- Artem Polyakov (NVIDIA)
- Aurelien Bouteiller (UTK)
- Brandon Yates (Intel)
- Brian Barrett (AWS)
- Charles Shereda (LLNL)
- Christoph Niethammer (HLRS)
- David Bernholdt (ORNL)
- Edgar Gabriel (UH)
- Erik Zeiske (HPE)
- Geoffroy Vallee (ARM)
- Howard Pritchard (LANL)
- Joshua Ladd (NVIDIA)
- Mark Allen (IBM)
- Matias Cabral (Intel)
- Nathan Hjelm (Google)
- Noah Evans (Sandia)
- Raghu Raja
- Ralph Castain (Intel)
- Scott Breyer (Sandia?)
- Shintaro iwasaki
- Xin Zhao (NVIDIA)
- New Person Trirage introduced himself.
- MPI Thread Level Discussion
- Two ways to initialize MPI: MPI_Init, and MPI_Init_thread
- Years ago we added an env var to MPI_Init to allow us to set env to request threading level, without having to modify tests.
- This env var isn't documented anywhere. Perhaps intended as an internal-only flag.
- Intentionally did not put this into MPI_Init_thread because you explicitly ask thread-level, so no env var check there.
- Recently in master, merged a commit to check this env var in MPI_Init_thread.
- This commit got merged to release branches (v4.0.x, and v4.1.x)
- What to do for v4.0.x and v4.1.x
- PR to revert in v4.0.x
- In v5.0.x and master - Issue 9332
- Do we want to make this uniform how
- If we want to expose to users, and not just magic/backdoor, should probably be an mca parameter.
- Discussion
- Would be nice for US for testing community.
- Nice to make it universal.
- Do we want to make it an mca parameter?
- mca paremter, move down to ompi_mpi_init level.
- Who do we talk to about Sessions [PR9097]
- If we do this uniformly, should just check with Session and make sure it's uniform.
- After mca system is standardized, but before mpi is initialized.
- is this a standardized arg to mpiexec? - George will check.
- On v4.0 and v4.1
- Jeff thinks it's a change in behavior (this var is set, and might get)
- Seen it multiple times where people.
- What if we use a different env var.
- Schedule: milestone is set for September for 4.0.7
- Howard is gone
- Jeff will look at v4.1.2 RC shortly.
- Schedule: milestone is currently (August) for 4.1.2 acculated bugfixes.
- A whole pile of PRs. Austen is testing a few to verify
- William has a PR he needs to backport today
- Went over the Github Project of [critical v5.0.x issues|https://github.com/open-mpi/ompi/projects/3]
- some progress on issues
- Described approach of rc1 on Sept 24, disabling any functionality that are blockers to allow for the rc.
- Worried that blockers might not be fixed in time, so will put in code to issue an error at runtime to prevent getting into those paths, and document it heavily.
- No discussion, so we're going forward with that.
- MPIAlltoallw needs to go in. Is a PR from Giles George
- Janjust - has a long outstanding one.
- Portals bugfixes incomming.
- https://github.com/open-mpi/ompi/pull/9326 should get into 5.0 too
- RMs will put in a disablement PR to prevent going down Issue 7830.
- If others can verify that this is an issue for them, perhaps it's site-local.
- No discussion
- MTT results look pretty good
- libevent - could only replicate this in his environment.
- No update
- Don't do the old system, use this new system for v5.0.0
- No discussion [Open MPI 4.0 API Compliance Github Project|https://github.com/open-mpi/ompi/projects/2]
- Looking okay.
- Ciscos results are still hidden by default.
- No discussion.