WeeklyTelcon_20180116

Open MPI Weekly Telecon

---

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

Attendees

--- Will fill out as meeting starts

• Geoff Paulsen
• Brian
• David Bernholdt
• Edgar Gabriel
• Geoffroy Vallee
• Jeff Squyres
• Howard
• Matthew Dosanjh
• Mohan
• Ralph
• Todd Kordenbrock
• Joshua Ladd
• Josh Hursey

Agenda

Minutes

Review All Open Blockers

Review v2.x Milestones v2.1.2

• Delayed until next week.
• No one has URGENT need, but would like to get this out

Review v3.0.x Milestones v3.0

• Schedule: RC2
  • On 3.x series trying to cut RCs on nightly tarballs.
  • Didn't get RC last week
  • Will get RC today.
• No Blockers on v3.0.x (one we JUST merged)
• Will Pull in PR4715
• Will Pull in PR4716
  • Issue 4563 - not seeing on little arm boxes here, Jenkins uses --disable-builtin-atomics.
• Comm Spawn - Documentation PR ready or pulled
• Issue 4509
  • We believe this is closed. Asked Nathan to close.

Review v3.1.x Milestones v3.1

• SCHEDULE:
  • Will shoot on getting Release Canidate out Friday.
• BLOCKER:
  • OSC monitoring fix (doesn't build with Portals 4)
    • PR4523
  • PMIx 2.1 PR4605
    • Ralph - there is cleanup issue with PMIx 2.1, but we have cleanup issues today
  • UCX one sided violating PR4688
  • Issue 4303
    • Probably just need to build a patch.

Review Master Master Pull Requests

• Issue PR4686
  • Jeff Tried to reproduce and failed.
  • Thought HCOLL was an issue, Artem took out, and put back.
  • Something going on in there. Possibly atomic related.
  • Might need Nathan's attention.
  • Someone could try reverting the one change to atomics to see if that caused it.
  • Mellanox will try to reproduce after reverting atomic change. Timing issue.
• Dynamic operations, a TON of sigfaults. All in opal_progress, during ompi_sync_wait multi-credit.
  • Something is wrong with atomics. Intercomm_create or Spawn.
  • Cisco is tickling the most, and will look at.
• PR4697 seems to have stalled. * Opal Progress change looks good for most interconnects. * TCP performance regression. * Pointer solution seems reasonable. * mellanox will try to implement pointer.
• Reg-ex expression creation.
  • PR4710
  • someone created a test and put it in make-check rather than MTT.
  • Then made the component static so that don't have to do make install
  • Dont think we should be adding tests to make-check
  • Question - Is there a Regex library we could use? Reg-ex is hard.
  • This is working pretty well, but did add Framework to allow for future components.

Process

• When your PR has been accepted into a release branch, please go to the issue, and remove the target of the release branch that it was just merged into. Attempting to automate this in the future.

MTT / Jenkins Testing Dev

• New Topic - We currently can't write unit tests against components.
  • Some way to say "this unit test is against this component".
  • Intel went through and did this internally for orte. Already hosted in public domain.
    • Ralph will send link to Brian to take a look.
• Python Client can't report back to database.
  • https://github.com/open-mpi/mtt/issues/614
  • Josh Hursey will look at.

Review Master MTT testing

Other Discussion

Next Face-to-face

• Probably looking at March or early April
  • San Jose or Dallas
    • Geoff will send out two Doodles for date and time.

Abandoning OpenIB BTL

• Discuss abandoning openib btl.
  • LNLL - is no longer paying anyone to maintain openib btl.
    • Nathan has a UCX BTL
  • ETA on GPU in UCX - basic minus CUDA IPC in test now.
  • Any warning message if on iWarp
  • What's the roadmap for this? 3.x or 4.x?

Oldest PR

Oldest Issue

Next face-to-face meeting

• pushed date to late feb or march.

---

Status Updates:

Status Update Rotation

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

---

Back to 2017 WeeklyTelcon-2017