Skip to content

Meeting 2020 07 10

Stephen Herbein edited this page Jul 11, 2020 · 1 revision

Agenda

Attendees

Josh Hursey (IBM)
Stephen Herbein (LLNL)
Ralph Castain (Intel)
Michael Karo (Altair)
Ken Raffenetti (ANL)
Karthik Vadambacheri Manian (OSU)
David Solt (IBM)
Martin Schulz (TUM)
Thomas Naughton (ORNL)

Notes

  • Q3 ASC Meeting
    • July 22. 11am - 5pm US Central
    • Discuss quarterly meeting dates for 2021
    • Link: https://github.com/pmix/pmix-standard/wiki/ASC-Q3-2020-Meeting
    • Agenda Summary:
      • Votes on 2 governance PRs and 2 standards PRs
      • V4 release update
      • Plenary from Storage WG
      • WG Updates
      • Technical and use case presentations
    • Q4 meeting will be virtual
  • Discussion on bringing wording changes into PMIx V4
    • Ralph: Changes being made in V4 are creating conflicts with open PRs that are being voted on. Nothing in those PRs are "contentious" in regards to V4, so it would be logistically fine to merge them into V4 rather than waiting until V5. Any knee jerk reactions against this proposal?
    • crickets
    • Ralph: we can discuss/vote in the quarterly meetings
    • Josh: the chapter 1 PR will most likely be voted in as is. You can always cherry-pick the commits from that PR onto your branch for V4 now
    • David: if the votes don’t pass, won’t it be difficult to pull them out in V5?
    • Ralph: yes, we would wait until they pass their second vote before pulling them into V4
    • Josh: target release date for V4: start of August?
    • Ralph: still the target, working on tools chapter now.
    • Josh: sounds like a draft at the end of July?
    • Ralph: yes, that sounds about right.
  • WG Updates
    • Stephen Herbein: Slicing/Grouping of functionality
      • Two big things since last meeting
        • Initial meeting of what turned out to be Power community, talking about interfaces and problems they have, w.r.t. Common interfaces and few pains to link against "yet another dependency", so possibly put into pmix to streamline. See discussion later in this agenda
        • Swaroop is working on MPI Sessions use case. Howard Pritchard & Dan Holmes talked about MPI-Sessions prototype and how they used PMIx, process-sets, process-groups and basis for MPI-Sessions.
        • Howard will join next few WG sessions to discuss
      • Still investigating best way to integrate code snippets. Looking into OpenMP’s examples/methods.
      • Added coverage checks for use-cases for interfaces and identifying gaps
    • Shane Snyder: Storage
      • Plenary planned for Q3 meeting
      • Defined attributes that would be helpful for querying storage information. Prototype implemented in OpenPMIx
      • Putting together a proposal for the new attributes. Hopefully will be a PR ready for voting at the Q4 meeting
    • Jai Dayal: Dynamic workflows
      • Jai has conflict with this timeslot
      • Ralph: continues to have interactions with workflow teams. Renewed push to put PMIx in kubernetes, possibly have it done by Jai/Intel. Josh will sync w/ Jai to coordinate efforts
    • Dave Solt: Implementation agnostic document
      • Working on Chapter 5: put, get, lookup, publish functionality
        • Polishing up introductions to the various sections
        • Making distinction between get/lookup and put/publish more clear
        • Should have a good number of non-contentious changes coming out of that quite soon
      • Plan is to continue forward onto following chapters once 5 is done
      • No changes were required to first PR that passed its first vote
      • Some changes to the second PR. Continuing to move that forward through the voting process
      • Work is 15-20% implementation agnostic work, and ~80% is cleanup/polishing of the document (this ratio was flipped for the first two chapters). Given these ratios, we are considering changing the name of the WG to be more broadly applicable
      • Ralph: there is a new open PR that addresses the reference of the pmix.org website within the standard for PMIx_Get attributes/keys. Would be helpful if Stephen and Dave can review.
      • David: WG meeting cancelled for this week due to some time off
      • Stephen: for the PR, are the keys that can be passed to PMIx_Get and their semantics included elsewhere is the standard?
        • Ralph: yes, they are
      • Ralph: the semantics for waiting for a non-reserved key and a reserved key are different. It is a bug in our code.
        • David: when you say it is a bug in our code, what is the intended semantics?
        • Ralph: Yes, he was passing the immediate flag to a Get request for a non-reserved key, but OpenPMIx was not respecting the immediate flag. The standard also does not clearly state that non-reserved keys will by default block.
      • Ralph: we (Jeff and I) have been doing video series on OpenMPI for the EasyBuild community, and an incredible number of people are engaging with the content. Maybe we should consider doing a PMIx series of a similar flavor where we walk through the semantics and usage of PMIx.
      • Josh: You hear PMIx a lot in the container community, but there is a significant percentage of people that view PMIx as an amorphous entity. Walking through the multi-version support and bootstrap will be good
      • Thomas: this is a great idea. Having short 10-15 minutes videos for OpenPMIx/PRRTE would be super helpful (e.g., past had similar experience with OpenStack), and can be more helpful than text sometimes.
      • Stephen: we can create use case documents from the videos
      • Ralph: we can also do the inverse: create videos from the use case docs
  • Discussion of Power API WG
    • Ralph: wanted to measure the temperature of the power community for some power interfaces in PMIx. After the initial meeting, it appeared that everyone had rallied around the PowerAPI, but after some further reaching out to the RM community, that doesn’t necessarily seem to be the case. The RM community feels caught in the crossfire and would want PMIx to create a higher-level abstraction. Initially, the early power APIs were very atomistic: setting and getting registers. Community wanted higher-level interfaces. There is a mismatch between what the power community is offering and what the RM community wants.
    • Ralph: Do we want to get into this layer of creating abstraction layer for Power? Do we even want to go this direction?
      • Martin: This seems much larger than PMIx, very complex problem and needs greater/broader involvement. Possibly includes PMIx, but may be larger. Seems wrong for a center wide power solution, maybe PMIx is not way to solve that?
      • Ralph: Would agree. Question being if PMIx could provide support in this space. But question being should we have a Power Working Group, or maybe we should have some representation/engagement into the other communities. Possibly have designated liaisons into other Power communities, to help see what PMIx could bring to assist.
      • Ralph: Can engage to see who might want to participate, Martin will also send contacts. Plan to just pass this off into that direction. Not sure PMIx should be the lead, but possibly have a role in that space. Martin agreed with that sentiment.
      • Josh: would be good if have liaisons in those groups, would be nice to have them give updates to PMIx community, e.g., status updates at quarterly meetings.
Clone this wiki locally