Skip to content

WeeklyTelcon_20170110

Geoffrey Paulsen edited this page Jan 9, 2018 · 1 revision

Open MPI Weekly Telcon


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

Attendees

  • Geoff Paulsen
  • Jeff Squyres
  • Artem Polyakov
  • Brian Barrett
  • George
  • Howard
  • josh Hursey
  • Josh Ladd
  • Ralph
  • Ryan Grant (SNL)
  • Todd Kordenbrock
  • Nathan

Agenda

  • Ralph put in the approved stuff this morning.
  • Still 7 PRs that need review.
  • No schedule yet.
  • Want to check that 2678 doesn't impact 1.10, but think it might.
  • 1.10.6 has an issue with patcher, but thoughts are not to drive a 1.10.6.
  • Will need another RC to fix issue 2678/2679 (master) not merged yet.
    • AMD save/restore all bits of register. Only happens with -fPIC and -O(2? or 3?)
    • will merge 2.0.2 of this after CI finishes.
  • Building for Ubuntu, had to set built in atomics for ppc64 to get FIFO to work.
    • Issue 2677 - But when we try atomics for ppc64 with -m32, make distcheck find another issue.
      • on ppc use a much faster algo. Nathan will look at.
    • Think we dealt with on Master, but may not have brought over.
      • Nathan added a check to see if 64bit was there. Saw similar on Intel 32bit virtual machine.

PMIx update

  • IBM Thinking is for Open MPI v2.1, would like move to PMIx 2.0
    • IBM is doing work for PMIx 2.0 component for OMPI v2.1.
      • IBM can support via external component
    • What does Open MPI want to do?
    • Need PMIx 2.0 to release before OMPI v2.1 if we choose to adapt.
    • Been running PMIx 2.0 on OMPI master for a while now.
  • Two steakholders for OMPI v2.1: IBM, Mellanox
    • IBM can use external component (working on doing that work now)
    • Mellanox prefers the PMIx v1.2.
  • PMIx v1.2.1 update - problem, we were overwriting

  • Still not getting email from AWS instance.
    • Cron email Brian every day that it ran.
    • Mail Authentication issue?
    • Should have an outgoing log from AWS instance.

MTT Dev status:


Exceptional topics

  • Progress thread in v2.0.x
    • ompi_info - always shows as off. Dead code down in BTLs, should we remove?
    • orte is locked to yes, ompi is locked to off.
    • not in v2.x series.
    • George - Right now only TCP supports an internal progress thread.
    • Issue is that OPAL_ENABLE_PROGRESS_THREADS is hard coded to 0 here and displayed to the user via ompi_info here
      • Since that is always displayed as no it can be misleading to the user.
      • There are a number of places in the code that still reference this variable, but are effectively dead code.
  • Still everything on track for January face to face?
    • Yes, wiki page updated with location.

Status Updates:

Status Update Rotation

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

Back to 2017 WeeklyTelcon-2017

Clone this wiki locally