Skip to content

Meeting 2022 06 09

Thomas Naughton edited this page Jun 9, 2022 · 5 revisions

Agenda

Attendance

Person Institution
Josh Hursey IBM
Kathryn Mohror LLNL
David Solt IBM
Isaias Compres TUM
Michael Karo Altair
Norbert Eicker JSC
Thomas Naughton ORNL

Notes

  • Administrative items
    • SC’22 BoF
      • This year looks to be back to normal levels so may be interesting to submit.  Discussion on who is attending (able to attend), whether BoFs will be streamed, etc.
      • Seemed like a good idea to submit. Josh & Kathryn will work on BoF submission.
    • PMIx Quarterly Meetings
      • PMIX ASC Quarterly Meeting: Aug 9 and 11, 2022
        • Announcement Deadline: July 12, 2022
        • Agenda Finalization: July 26, 2022
      • PMIx ASC Monthly Meeting:
        • Thursday, July 14, 2022
  • PMIx v4 update
    • Plan for mid/latest 2022
  • PMIx v5 update
    • Making progress, no planned changes to timeline
    • ABI is main item and making progress (no known issues currently)
  • Working Group updates
    • Implementation agnostic document: Dave Solt
    • Tools: Isaías A. Comprés
      • OpenPMIx going to be big part of EU projects
      • Next steps to implement negotiation protocol between SLURM and MPICH with PMIx, looking into the malleability – will evaluate if PMIx interfaces are sufficient to support these needs.  Could be extensions to PMIx library or as a separate library, but looking into details/plan for this use case and will see what works best.
    • Dynamic workflows: Justin Wozniak
      • Not much traffic lately
      • TODO - Need to reach out and check status
  • ABI Update (Josh)
    • https://github.com/pmix/pmix-standard/issues/365
    • Added support in OpenPMIx and updated the ABI example in pmix-abi repo (see ticket for links).  Demonstrate the ability to quarry for ABI_VERSION info.
    • Also working on way for users to know provisional vs stable distinction on pieces w.r.t. Versioning for supported functionality.
  • PMIx Publish Discussion (David)
    • As looking at publish, considering a new attribute to separate things out.
    • As part of this, recognized there’s a lack of guidance on handling duplicate directives, e.g., two calls to pmix_get with two timeouts, unclear which should be used from specification perspective.  Underspecified on expected behavior.
    • Suggestion: Maybe look at what openpmix does and reflect that into spec. Having something defined for behavior seems better than current ambiguity.
    • Q: Do we know of any apps that pass duplicate keys and expect the behavior currently done w/ openpmix?  (none mentioned)
    • Possibly put this item on todo and avoid blocking current work w/ publish/lookup text changes
Clone this wiki locally