Skip to content

ASC Q2 2021 Meeting

Aurelien Bouteiller edited this page May 14, 2021 · 12 revisions

PMIx Standard Administrative Steering Committee (ASC) 2Q 2021 Meeting

Quick Links

  • Governance Document [latest]

Agenda (Finalized on April 27, 2021)

This meeting has a floating agenda with specific synchronization points to keep us on track. Rough time estimates are provided per agenda item, and the co-chairs plan to cover the topics in the order seen below. However, since some agenda items will take longer/shorter than anticipated an exact start/end timing is not guaranteed and some items may float to the second day. If you are not able to attend the full meeting and are presenting then please let the co-chairs know and we can plan accordingly.

Day 1: May 11 (10 am - 1 pm US Central Daylight Time)

Start End Topic
10:00 am 10:05 am Gathering (Josh)
10:05 am 10:10 am Roll Call (We will start roll call promptly at this time)
10:10 am 11:30 am Discussion of agenda items
11:30 am 11:45 am Break
11:45 am 1:00 pm Discussion of agenda items

Day 2: May 13 (10 am - 1 pm US Central Daylight Time)

Start End Topic
10:00 am 10:05 am Gathering (Kathryn)
10:05 am 11:30 am Discussion of agenda items
11:30 am 11:50 am Voting and Break Doodle Vote Link
11:50 am 12:30 am Administrative and Working Group agenda items
12:30 am 12:45 pm Technical and Use Case Presentation(s)
12:45 am 1:00 pm Closing discussion and wrap up

Agenda Items

Administrative and Working Group Agenda Items

  • Review 2021 quarterly meetings dates and plans
3Q 2021 (July 1 - Sept. 30) - Virtual
 - 2 day: Tues., July 20 & Thurs., July 22

4Q 2021 (Oct. 1 - Dec. 31) - Virtual
 - 2 day: Tues., Oct. 26 & Thurs., Oct. 28
  • ASC Membership
    • Vote on new ASC Members (None)
    • Call for new ASC Members
  • Release Planning
  • Working Group Updates (~ 10-15 minutes each)
    • Client Separation / Implementation Agnostic Document
    • Slicing/Grouping of Functionality
    • Dynamic Workflows
    • Storage
    • Tools
    • Open Call for New Working Groups
  • Technical and Use Case presentations
    • Role of PMIx with Containers in HPC Environments (Josh ~ 20 min)
  • Additional discussion items

Meeting Notes:

Attendance

Person Institution Day 1 Day 2
Josh Hursey (IBM) Day 1 Day 2
Aurelien Bouteiller (UTK) Day 1 Day 2
Swaroop Pophale (ORNL) Day 1 Day 2
Brian Barrett (Amazon AWS) Day 1
Stephen Herbein (LLNL) Day 1 Day 2
Howard Pritchard (LANL) Day 1 Day 2
Kathryn Mohror (LLNL) Day 1 Day 2
Ken Raffenetti (ANL) Day 1 Day 2
Thomas Naughton (ORNL) Day 1 Day 2
David Solt (IBM) Day 1 Day 2
Michael Karo (Altair) Day 2
Isaías A. Comprés (TUM) Day 2

Day 1: May 11, 2021

Introduction / meeting schedule

Roll call

  • First time: Amazon AWS

PR #328 (Stephen Herbein): appendix with MPI wire-up use case and tools use case

  • no major changes since last votes, just a rebase
  • No questions

PR #349 (Swaroop Pophale): appendix with MPI Sessions use case

  • Reading
  • No questions

PR #327 (Josh Hursey): clarify roles of consumers of the PMIx interface

  • Reading
  • Q: what about “all-roles”; discussion settled in favor of removing all-roles in favor of fully enumerated roles, revision vote will be held tomorrow

PR #347 (Dave Solt): chapter 2 changes

  • Review changes from read previously (last year)
  • TODO: insert link to Dave slides
  • Note - dave’s footer says asc q1-2020 meeting, likely initial reading date and read again 5/11/21
  • “Pmix_” (slide 15 of presentation) shall it apply to attribute string values? No because legacy users could have PUT pmixABC key strings and check them only with the first 4 chars. It is unlikely someone has done that, but we can’t be sure.
  • “Pmix” on slide 16? Should it be “pmix_”; yes
  • Does the new text disallow that clients can create “pmix_xyz” attribute strings? Yes, we should clarify that the clients/tools cannot do that. There is a technical problem because servers use the client interface; and servers should be able to do this? It causes potential namespace conflict problems, but pmix_ is the way we recognize predefined server keys (which have a different behavior); we have not resolved that greater issue yet; we will need to come back to it later.
  • Suggested change (to be added on ticket) fix the symbols problem, defer the string problem to later: Users shall not use the \textbf{\code{"PMIX_"}}, \textbf{\code{"PMIx_"}}, or \textbf{\code{"pmix_"}} prefixes for symbols in their code so as to avoid symbol conflicts with \ac{PMIx} implementations. Revision exception vote tomorrow

Governance PR #29: (Josh Hursey) errata changes process

  • Reading
  • No questions

Governance PR #30: (Josh Hursey) major text changes

  • Reading
  • No questions

Governance PR #32: (Kathyrn Mohror) Modify text for formal vote on provisional items

  • Reading
  • No questions

PR #332: (Brian Barrett/Ralph) Four missing data management functions

  • Reading
  • Const change will go as revision exception
  • in/out reordering; Ken to propose another param order; we will need to update (unreleased) openPMIX 4.1 and PRTE 2.0
  • Revision exception vote; then provisional vote

PR #334: (Brian Barrett/Ralph) Add a data type constant for pmix_nspace_t

  • Reading
  • No questions

PR #331: (Brian Barrett/Ralph) Add a few simple, but helpful macros

  • Reading
  • XFER comment1 (see ticket); it;s a simple copy
  • XFER comment2 (see ticket); should it be XFER_PROCID or PROCID_XFER?
  • Aurelien will follow-up with PR suggestion, note to also include “d,s” for macro args for clarity

PR #335: (Brian Barrett/Ralph) Provide access to resource usage for processes and nodes

  • Reading
  • Comments related to descriptions for fields (see ticket)
  • Should the reports be in kB? Size_t? Stay as float MB? Size_t should be enough and precise, but less portable than float (if reading data blob from another machine w/ different architecture); Having a structure could ‘lock-in’ current fields, some people expressed that having accessors could be better
  • Comment on ticket : can decorate with RANGE_SESSION etc; people agreed
  • No implementation for some attributes in OpenPMIX Need more info about the next step; will discuss it with Ralph Castain before thursday if possible

PR #325 (Biran Barret/Ralph) stale language

  • Reading
  • No questions on content
  • Moving the V4.0 to V4.1 should be another PR

Day 2: May 13, 2021

PR #31 (Josh Hursey): clarify there are 2 secretaries

  • Reading
  • No questions

New ASC members call

  • Nobody
  • (maybe Brian Barret would be interested, but he couldn’t have a representative today; Josh to reach out)

Voting @11:30 CT

  • Revision exception items from Tuesday
  • PR #331: we had verbal agreement from Brian on Tuesday on the proposed changes, we will move forward with this ticket
  • PR #327, PR #332

Standard release discussion v4.1 (Josh)

Standard release discussion v5.0 (Ken)

  • Timeframe: sometimes in 2022
  • WG Use cases: server-side for business cards and tools; client-side for storage, workflows, version negotiation
  • WG Implementation agnostic; return codes; 3 roles of consumers; const?; chapter reorg (5, 6, 7)?
  • WG Storage; nothing additional from the already accepted provisional PR
  • WG Dynamic workflows: TBD
  • Ken would like people that have strict deadlines (deliverables for particular feature, etc) to reach out; we have a time-based release cycle, but we like to know how much wiggle room we have with fudging the schedule; full agreement that we should prefer short-cycle releases, even if we don’t push out -all- of the new stuff in the release, better some than none.

Updates from WG

Impl. agnostic (Dave)

  • Still working on main thrust, removing things that are implemementation specific and general clarification;
  • Chapter 5 is almost ready;
  • Work on 6-7 have started and will continue in near future. Dave estimates the WG is halfway through, and hopes it can be finished this year.
  • Macros for return success or something else (e.g., non-blocking calls) PR on chapter 2 for namespacing is also moving forward

Use-cases (Stephen)

  • Q1 completed appendix for bootstrap, debuggers/tools, hybrid prog models (voted in)
  • Q2 created PR for MPI Sessions use case
  • Q3 lookup for new use cases, extending use cases for server-side API; update existing use-cases; collab with storage WG to design a use-case
  • Meet @ Thursday 9am PT

Dynamic workflows (Josh summarized activities)

  • Discussing Savanah and Tuda(?), ORNL workflow projects. Looking at how to use PMIx groups and sessions in workflow context;
  • Kubernetes were discussed before; prototyping activities The group is more focused on how to use existing APIs in workflow context, and identify potential missing API (there were discussions about having some GPU attributes, join the group to know more)

Storage WG (Shane)

  • Q1: query accepted
  • Since then attendance has been a bit low to capture enough use-case/user stories
  • Maybe put the WG on the backburner and see if users call out attention on exciting new issues related to storage; WG will keep an eye on provisional PR and followup if needed

Tools WG (Isaias Compres)

  • Tool portability with PMIx
  • Many tools hook directly into MPI runtime; but some measurements require root access, which is not appropriate for user-code; could be delegated to the PMIx server (system-code)
  • Investigate low-overhead interface (in contrast to acceptable overhead interface) for high-precision measurement; many of PMIx interfaces are already low-overhead, but systematic approach needed

Discussion on how to help implementation reqs for Open PMIx standardization

  • Secure additional funding to support implementation effort?
  • Existing support from
    • TUM/Julich
    • DOE OMPI-X

Call for new WG

  • No new WG proposed

Technical presentation: Josh Hursey; Role of PMIx with Containers in HPC Environments

Closing words

Clone this wiki locally