-
Notifications
You must be signed in to change notification settings - Fork 859
WeeklyTelcon_20161004
Geoffrey Paulsen edited this page Jul 25, 2023
·
2 revisions
- Dialup Info: (Do not post to public mailing list or public wiki)
- Geoff Paulsen
- Jeff Squyres
- Brad Benton
- David Solt
- Edgar Gabriel
- Howard Pritchard
- Josh Hursey
- Joshua Ladd
- Nathan Hjelm
- Ryan Grant
- Sylvain Jeaugey
- Todd Kordenbrock
- Milestones
- 1.10.4
- Still Nothing new. No one raised any issues at this time.
-
- 2.1.0
- Sept 30th deadline for New Features. Jeff sent email last Wed or Thursday.
- Mpool Rcache re-write PR2101. Nathan reverted all of the spot fixes, and then applied all of them.
- Nathan has 2 more PRs for trivial features he'd like to add.
- Add flag enumerator to mca base - one of the cherry-picks will be much harder if it goes in before.
- Been on master for many months, but many many cherry-picks, so PLEASE review.
- Affects every BTL, been couple of different
- Goals: clear interface between mpool and rcache. supports memkind.
- orthogonal to memhooks, because only affects internal allocations.
- confusing, because it used to be very confusing, but NOW is separating this out. All explicit, allocations, and then separate registrations.
- Nathan has 2 more PRs for trivial features he'd like to add.
- C++ wrappers for OSHMEM - Failed Jenkins but passes by hand. Resolve before merging.
- PMIx 2.0
- Progress with Artem and Josh - Changed directions to use 1.1.2 + dstore shared memory.
- probably not end of the week.
- Vader out of order issue (single threaded case):
- Relying on OB1. When you setup fastbox, everything must go through fastbox. When fastbox is full, everything goes on a queue to be processed later.
- Fix could be to only preserve a fastbox if nothing on pending list... but that causes performance implications.
- Easy enough to fix, can make 2.1.0
- No expected effect on vader multi-threaded.
- Vader multithreaded - All we care about is threading in the context of a thread. Not for 2.1.0. This will require a redesign.
- May be research.
- Ordering with multiple threads means something different than in single thread.
- Sequence numbers (Per MPI_Send()) are problematic in multi-threaded modes.
- If you have multiple routes, you need sequencing.
- What to do in multiple thread case... needs to be pushed to a redesign.
- In Master can use PMIx 2.0 with external. In Master, internal component has already upgraded to 3.0.
- IBM and Mellanox, along with Nathan and Howard (LANL) will meet to discuss getting this work done quickly for OMPI 2.1.0.
- 2.1.0
-
GITHUB Tags / Milestones - not well documented in wiki.
- push-back tag - means it's gone back to the author.
- Other tags have changed meanings over time.
-
Discussion on Giles OSHMEM - onexit PR2121
- Seems okay, not sure what the purpose of it.
- Need to ask Giles why he wants this.
- Giles said on PR2120 that it was addressing a corner-case (--enable-static or --disable-visibility).
-
SPI - http://www.spi-inc.org/
- getting people to approve of these.
- We'll be on Oct 12th Agenda. Once they formally invite us, then we have 60 days to agree / decline.
- Works solely on a volunteer basis, so very inexpensive.
- End of September for soliciting feedback on using SPI.
- Open MPI will hold a formal vote after we receive the formal invite (in mid-to-late-December?)
- Tennessee owns the logos, but they will recreate with meta-data giving them to open source.
-
New Contribution agreement / Consent agreement / Bylaws.
- Will need a formal vote by members.
- End of October for discussion of new contributor agreement / bylaws.
- After that we'll set a date for voting.
- New Contributor agreement
-
EuroMPI 2016 In Edinburgh - Spet 25-28
- MPI Forum: Sept. 21-23
- People might be traveling next week
Review Master MTT testing (https://mtt.open-mpi.org/)
- Date of another face to face. January or February? Think about, and discuss next week.
- Geoff will put up a doodle to solicit input.
- LANL, Houston, IBM
- Cisco, ORNL, UTK, NVIDIA
- Mellanox, Sandia, Intel