WeeklyTelcon_20160322

Open MPI Weekly Telcon

---

• Dialup Info: (Do not post to public mailing list or public wiki)

Attendees

• Jeff Squyres
• Geoff Paulsen
• Brad Benton
• Geoffroy Vallee
• Howard Pritchard
• Josh Hursey
• Nathan Hjelm
• Nysal
• Ralph Castain
• Sylvain Jeaugey
• Todd Kordenbrock
• Tommy Janjusic (Mellanox)

---

Agenda

Review 1.10

• Milestones: https://github.com/open-mpi/ompi-release/milestones/v1.10.3
  • PR1004 - no INPLACE for neighborhood. Ready?
  • PR1030 - needs to be rebased to pull in.
• 1.10.3 in April - collecting some smaller fixes accumulating in the branch.
  • Nothing critical at the moment

Review 2.0.x

• Wiki: https://github.com/open-mpi/ompi/wiki/Releasev20
• Blocker Issues: https://github.com/open-mpi/ompi/issues?utf8=%E2%9C%93&q=is%3Aopen+milestone%3Av2.0.0+label%3Ablocker
  • Issue 1406
    • TCP BTL THREAD_MULTIPLE deadlock
      • If old and new rewrite of TCP BTL are "compatible", then we can switch based on threaded state of MPI_Init.
      • OR could require two different TCP components "tcp" / "tcpmt", and expose this issue to users.
    • George working on patch.
  • Is Issue 299 an issue on 2.0.0? If so, that should also be a blocker!
    • Issue linked on mailing list. User sent
    • Related to Issue 429?
      • Would merging opal and orte into one lib workaround this? - Probably, but that's alot of work.
    • Still issue with UCX (who does hooking?)
    • If we make a blocker, can accelerate fix? - Yes.
    • Some assembly required. Might need someone to
    • Assign to Mark Allen. - HIGH PRIORITY.
  • Unblock PR 1353 move to 2.1?
    • For 2.0 document what behavior is, and push change back to 2.1.
  • 1038 - waiting on review mellanox
  • 1015 - segv ibm - Howard
  • 1014 - don't return Err_pending from collectives?
  • Jeff will test USOCK today. - Coverity found an issue, Ralph will look at tomorrow.
  • George has a patch to fix TCP (smaller patch to fix
• Milestones: https://github.com/open-mpi/ompi-release/milestones/v2.0.0 *

Review Master?

• Master tests are failing.
• C++ bindings failing to compile is a symptom. cyclic dependencies?

MTT status:

• Working on submit interface to make client easier to submit to. Should help Ralph's team.

• PR1483 - allows base btl iterators to determine flag values.

  • Only issues is how this is exposed to MTT
  • Expose to Tools Working Group
  • Gives you comma delineated list, and gives you all possible values.

• PR1482 - Support in MCA base to force components to always be on.

  • If you make your component static, and component says it's Always-on.
  • Key usage for this is forcing BTL self (If you say not-self, will give error and abort)
  • Useful for hook licensing framework.
  • need to look at other components (Maybe Coll-basic), (Same with libnbc)

Status Updates:

• Mellanox -
• Sandia - Todd Kordenbrock - Made a PR1443 PR 1037 for rondevue
• Intel - Working on Debugger Attach. If Ralph simulates debugger attach it works okay.
  • Ralph doesn't have partner license.
  • Not sure what the issue is, does mpirun not know if it's being debugged? Is the message not getting through?
    • How does this relate to USOCK issue?
    • Ralph was hoping to fix this, and debugger attach, and USOCK
    • Once John helps Ralph reproduce issue, can do USOCK this right away.
  • Still working alot on PMIx event reservation system.
  • Still working on ORTE launch fun.

---

Status Update Rotation

1. Cisco, ORNL, UTK, NVIDIA
2. Mellanox, Sandia, Intel
3. LANL, Houston, IBM

---

Back to 2016 WeeklyTelcon-2016