WeeklyTelcon_20151208

Open MPI Weekly Telcon Minutes 12/8/2015

---

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

Attendees

• Todd Kordenbrock
• Edgar Gabriel
• Geoffrey Paulsen
• Geoffrey Vallee
• Ralph Castain
• David Solt

---

Agenda

Review 1.10

• Milestones: https://github.com/open-mpi/ompi-release/milestones/v1.10.2
  • a bunch of PRs open against master, and some of them may need fixed on 1.10.x
  • Detecting failures None of the MTT we run nightly, are running MPICH at night.
  • We should add MPICH to nightly runs
• 1.10.2 Release Canidate anticipating week of 12/14 - 12/18
  • Integrate PR #811?

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
  • 1064 - Ralph / Jeff, is this do-able by december?
  • Dynamic add procs is busted now when set value to 0 (not related to PMI-x)
    • A bunch of paths, that may fix this... don't know yet.
  • External Linage issue
    • Distros won't accept Open MPI without external linkage to ?
      • first time we have a part of Open MPI that has an external linkage to libevent.
    • in the past the App was using libevent (version different than internal usage)
      • Solved by renaming symbols in internal version.
    • Problem now is that we have another part of Open MPI (PMPI-x) that also links against libevent.
    • If we remove embedded version of libevent, that would be okay.
    • it's when we build against internal version of libevent, then PMI-x ???
    • One solution is to no longer package libevent with Open MPI.
    • Reason we package it, is so that people don't have to download libevent along with Open MPI.
      • Now libevent-pthread is available on most intros.
    • We DO require libevent to be thread enabled. and check for this in configuration,
      • No performance implications.
  • Debugger support is broken on Master and on 2.x
    • Ralph needs version PMIx 1.2 to do it.
    • He thinks he can come up with a solution in the January time frame.
    • If we start to use PMI-x for the debugger, then the proc could tell a resource manager that they want to be debugged.
• Milestones: https://github.com/open-mpi/ompi-release/milestones/v2.0.0
  • One of us will go through ALL Issues for 2.0.0 to ask if they can be moved out to future release, or apply blocker status.
  • RFC on embedded PMIx version handling
  • RFC process wiki page?

MTT status:

• should add MPICH tests to nightly tests to improve
  • still a lot of failures on nvidia test cluster... still in transition.
  • Couple of errors on Jeff's cluster, not worth worrying about yet in Comm Dup stuff. KNOWN
  • Geoffrey Vallee has a cluster that is almost ready to add. Right now it's just -in trial.
    • Geoffrey found
    • next year will add new clusters in lab.
  • Ralph has a new cluster that he's working to get added to Jenkins.

Status Updates:

• LANL - Not Present
• Houston
  • Edgar ran a number of performance numbers.
  • Also ran into Issue #1191. Since Mid November. Issue #1191
• HLRS - Not Present
• IBM -
  • working on: supporting multiple versions of LSF at runtime (rather than build time)
  • setting up internal git mirror.
  • running internal code scans on master branch.
  • Dave's ompi_comm_info work
  • like to use new RTC process for init / finialize vendor components (licensing)

---

Status Update Rotation

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

---

Back to 2015 WeeklyTelcon-2015