Skip to content

ASC Q1 2021 Meeting

Jump to bottom
Thomas Naughton edited this page Mar 13, 2021 · 21 revisions

PMIx Standard Administrative Steering Committee (ASC) 1Q 2021 Meeting

Quick Links

Agenda (timeline in the chart below)

  • Review 2021 quarterly meetings dates and plans
2Q 2021 (April 1 - June 30) - Virtual
 - 2 day: Tues., May 11 & Thurs., May 13

3Q 2021 (July 1 - Sept. 30) - Virtual
 - 2 day: Tues., July 20 & Thurs., July 22

4Q 2021 (Oct. 1 - Dec. 31) - Face2Face if possible
 - 2 day: Tues., Oct. 26 & Thurs., Oct. 28

Agenda Timeline (Finalized on Feb. 2, 2021)

We will try to keep to this timeline as best as we can. However, discussion items may take longer/shorter than anticipated and as a result, the agenda may need to be adjusted during the meeting.

We will start roll call promptly at 10:15 am on Feb. 16. After this point, the co-chairs may decide, during the meeting, to adjust the timeline based on the discussion.

All times in US Central. (Last Update: Feb. 2, 2021)

Feb. 16 (10 am - 1 pm US Central Daylight Time)

Start End Topic
10:00 am 10:05 Gathering (Josh)
* Slides
10:05 am 10:15 Discuss 2021 quarterly meetings
10:15 am 10:25 Roll Call, Vote on New ASC Members, Call for New ASC Members
- Roll call
- Call for New ASC Members
10:25 am 10:35 Governance PR Second Vote
* Clarification of timeline for quarterly meetings items
- https://github.com/pmix/governance/pull/17
* Introduce Revision Exception Vote
- https://github.com/pmix/governance/pull/18
Voting Link
10:35 am 11:05 PMIx Standard PR Reading: Storage Working Group
* Storage Query Support for provisional acceptance
- https://github.com/pmix/pmix-standard/pull/280
11:05 am 11:50 PMIx Standard PR Reading: Functionality Slices Working Group
* Use Case Appendix for provisional acceptance
- https://github.com/pmix/pmix-standard/pull/328
11:50 am 12:05 Break
12:05 pm 12:10 (Announce Voting Results)
12:10 pm 12:15 PMIx 4.0 Released on Dec. 22, 2020
12:15 pm 12:25 PMIx 4.1 Release Planning
12:25 pm 12:35 PMIx 5.0 Release Planning
* v5 Slides
12:35 pm 1:00 Additional Discussion Items

Feb. 18 (10 am - 1 pm US Central Daylight Time)

Start End Topic
10:00 am 10:05 Gathering (Kathryn)
* Slides
10:05 am 10:35 PMIx Standard PR Reading: Storage Working Group
- https://github.com/pmix/pmix-standard/pull/280
10:35 am 11:05 PMIx Standard PR Reading: Functionality Slices Working Group
* Use Case Appendix for provisional acceptance
- https://github.com/pmix/pmix-standard/pull/328
Voting Link
11:05 am 11:30 Plenary: Governance Items (Josh Hursey)
* Governance: Major Descriptive Text Changes Process
- https://github.com/pmix/governance/issues/21
* Governance: Errata Changes Process
- https://github.com/pmix/governance/issues/20
11:30 am 11:45 Plenary: Technical Items (Ralph Castain)
* https://github.com/pmix/pmix-standard/pull/325
* https://github.com/pmix/pmix-standard/pull/331
* https://github.com/pmix/pmix-standard/pull/332
* https://github.com/pmix/pmix-standard/pull/334
* https://github.com/pmix/pmix-standard/pull/335
* https://github.com/pmix/pmix-standard/pull/333 (Withdrawn)
11:45 am 12:00 Break
12:00 pm 12:05 (Announce Voting Results)
12:05 pm 12:35 Plenary: Rules for non-standardized keys (Dave Solt)
12:35 pm 12:55 Working Group Updates
- Client Separation / Implementation Agnostic Document Working Group
- Slicing/Grouping of functionality Working Group
- Dynamic Workflows Working Group
- Storage Working Group
* Open Call for New Working Groups
12:55 pm 1:00 Additional Discussion Items

Day 1: Feb. 16, 2021

Day 1: Attendees

Josh Hursey (IBM)
Aurelien Bouteiller (UTK)
Brice Goglin (Inria)
Thomas Naughton (ORNL)
Stephen Herbein (LLNL)
Kathryn Mohror (LLNL)
Ken Raffenetti (ANL)
Martin Schulz (TUM)
Nat Shineman (OSU)
Swaroop Pophale (ORNL)
Ralph Castain (self/Nanook)
Shane Snyder (ANL)
John DelSignore (Perforce/TotalView)
Michael Karo (Altair)
Jai Dayal (Intel)
David Solt (IBM)

Day 1: Notes

  • General opening statement
  • Voting
  • Planning for Q2-Q4 meetings; Q4 tentatively will be f2f; dates in the slide deck and agenda
  • New members:
    • Ralph Castain has changed employer, looking how he can participate as an independent contractor (Nanook Consulting)
    • Motion to fast-track Ralph given past participation history, no opposition
    • Note: Reach out to Jai D. at Intel about representation/etc.
    • Note: New OSU attendee - Nat Shineman will be attending in future (OSU already has voting status)
  • Roll call
    • Altair (no, arrived after roll call)
    • Argonne (yes)
    • IBM (yes)
    • INRIA (yes)
    • Intel (no, arrived after roll call)
    • LANL (no)
    • LLNL (yes)
    • Nvidia/Mellanox (no)
    • ORNL (yes)
    • OSU (yes)
    • TUM (yes)
    • Pathforce (no)
    • UTK (yes)
    • Nanook (yes)
  • Governance Doc (2nd reading) - https://github.com/pmix/governance/pull/17
    • Clarifying language in governance doc for items to be ready in advance of quarterly meeting.  4 Weeks in advance of meeting if plan to read at meeting, 2 weeks prior to meeting, items can be changed w/ feedback but frozen thereafter up to meeting
  • Governance Doc (2nd reading) - https://github.com/pmix/governance/pull/18
    • Revision exception vote for PRs - can choose to allow changes after the “freezing” period (e.g., after first vote), introduce minor changes and assuming all vote “yes” then item can continue as normal (i.e., like MPI Forum’s “no-no” vote)
  • Voting link Voting Link
    • Second Vote: Governance PR #17 -> Passed
      • 9 yes / 0 no / 0 abstain / 5 no present
    • Second Vote: Governance PR #18 -> Passed
      • 9 yes / 0 no / 0 abstain / 5 no present
  • Provisional item reading - https://github.com/pmix/pmix-standard/pull/280
    • #280 - Storage system query support
    • Discussion asked about using string as constant, a new scenario. Previously all constants are numeric (e.g., enum).
    • Could possibly use numeric constant and helper to convert num to string, but the intent is to have human readable “label” for storage system with the query of the filesystem.
    • The use of strings (e.g., “unify”) would have less potential for collisions with the names in contrast to using purely numbers that might be defined in to places.
    • Could always have a collision, only can control collisions with standard’s specification of the info (integer or string)
    • Main motivation to use integers instead of strings is for bandwidth/storage when sending information around, arbitrarily long strings could use more space.
    • Need to have clear statement of what is reserved and what is arbitrary and can be user-defined (“wild west”), whether that be numeric or string.
    • Q: Any strong objection to staying with strings?
      • No conclusive response during meeting
    • Purpose of this to differentiate the file-system, so users can configure based on the fs type
    • The name is just start of info will be getting back from query, later will open up to have more details.
    • Some discussion about accessibility types, specifically  PMIX_STORAGE_ACCESSIBILITY_{CAMPUS,BUILDING}, possibly condense those into PMIX_STORAGE_ACCESSIBILITY_FACILITY
    • TODO - Items to take back to consideration
      • Storage type (number/string) needs to be discussed/revisited
      • Accessibility CAMPUS/BUILDING to FACILITY
      • Minor typos
    • TODO - time on Thursday for follow-up on this item
  • Provisional item reading - https://github.com/pmix/pmix-standard/pull/328
    • #328 - Appendix with use case descriptions
    • First reading
    • Some typos/style review comments added to PR
    • Comment - good to add “co-launch” attributes on debugger use case to show how to launch debugger on appropriate node when attaching to target proc.
    • Comment - sync w/ MPI Sessions WG related to tool interaction
    • Martin remarks that programming models term is abused in the programming model-interaction section (around pp 542-543); he proposes to help fix the text to clarify that it is coordination between implementations of various programming models.
    • Martin notes that OpenMP is trying to get rid of master-thread; we may need to double check what is the new terminology.
  • PMIx 4.0 Release - https://github.com/pmix/pmix-standard/milestone/3
    • Done! (Dec. 22, 2020)
  • PMIx 4.1 Release - https://github.com/pmix/pmix-standard/milestone/8
    • Planned for Spring 2021
  • PMIx 5.0 Release - https://github.com/pmix/pmix-standard/milestone/6
    • Release managers: Ken R. (ANL) and David S. (IBM)
    • TODO - add link to Ken’s slides

Day 2: Feb. 18, 2021

Day 2: Attendees

Josh Hursey (IBM)
Thomas Naughton (ORNL)
Brice Goglin (Inria)
Kathryn Mohror (LLNL)
David Solt (IBM)
Ken Raffenetti (ANL)
Michael Karo (Altair)
Nat Shineman (OSU)
Swaroop Pophale (ORNL)
Ralph Castain (Nanook)
Shane Snyder (ANL)
Howard Pritchard (LANL)
Stephen Herbein (LLNL)
John Delsignore (Perforce)
Aurelien Bouteiller (UTK)
Howard Pritchard (LANL)
Martin Schulz (TUM)

Day 2: Notes

  • Introductions
  • Storage Working Group - https://github.com/pmix/pmix-standard/pull/280
    • Shane Snyder
    • Review of follow-up edits from Tuesday reading of proposed provisional changes
    • Note - Decided to have a general string type, e.g., “PMIX_STORAGE_TYPE” as a string and that will contain “gpfs”, “ext4”, etc. as return from statfs
      • Comment - probably useful to have Advice to Implementers to follow strings from statfs for what those strings will contain to help middleware developers. As a future improvement
      • Comment - there is a similar item to this in the fabric API, so possibly to have both of those in future to get that info
      • Note - the kernel community does standardize those strings, so if use the values returned there, we’ll be consistent across platforms.  So thinking PMIx does not need to standardize those things, but leverage others
    • There was a minor change in advice to impl., removing topologies from storage_medium_t; no comment from the assistance on that change
    • Removed PMIX_STORAGE_ACCESSIBILITY_{BUILDING,CAMPUS,WAN} and replace with PMIX_STORAGE_ACCESSIBILITY_REMOTE
      • Comment - Suggested future improvement - refine text about accessibility “CLUSTER”, phrase accessible from where the job is running.  Example, center-wide filesystem spanning multiple cluster, the intent would be that the fs is accessible from where the job is running (cluster).
    • Revision exception procedure vote will be taken on this today
  • Functionality Slicies Working Group - https://github.com/pmix/pmix-standard/pull/328
    • Stephen Herbein
    • Review of follow-up edits from Tuesday reading of proposed additions use-case appendix https://github.com/pmix/pmix-standard/pull/328
      • Comment - Revisit the pthread as “programming model” to an API, possibly expand in future to have more broad threading model edit.  For now just change to posix threading in example.
  • Voting link for #280 and #328 - https://doodle.com/poll/x3wztdfxvnna4n9c
    • All passed
    • Revision Exception: Standard PR 280 -> Passed
      • 10 yes / 0 no / 2 abstain / 2 no present
    • Revision Exception: Standard PR 328 -> Passed
      • 11 yes / 0 no / 1 abstain / 2 no present
    • First Vote: Standard PR 328 -> Passed
      • 11 yes / 0 no / 1 abstain / 2 no present
  • Plenary - Governance items
    • #21 - Major Descriptive Text Changes Process
      • https://github.com/pmix/governance/issues/21
      • Discussion centered around how this process may cause heavy ongoing merge burden, or merge rush when preparing a major release
      • We don’t see a better way to deal with it
        • Release manager will have to do the merging when that happens
        • Release manager will have to raise a flag when merge conflicts raise issues that need further ‘major text changes’
        • Removal of provisional items causing conflicts will have to be handled on a case-by-case basis when they cause issues in other provisional items or major text changes; we do not wish a formal process for this, we will deal with it by creating tickets/pr when the issue arises and follow procedurals to integrate them.
    • #20 - Errata Changes Process
      • https://github.com/pmix/governance/issues/20
      • Modeled after provisional, adding a vote. Note, errata reading deemed too large for ASC could be moved to a different process
        • Comment - having vote seems good b/c otherwise no way of knowing change event.  Also suggestions on rapid release good.
        • No concerns mentioned from attendees
  • Plenary - Technical items
  • Plenary - Rules for non-standardized keys
    • Dave Solt
    • TODO - add link to slides
      • Comment - setting an attribute for custom keys allows it to behave (e.g., not block) when using non-standardized keys.  Just a way to have behavior similar to standardized keys.  (i.e., optional/immediate to add timeout attribute)
      • Comment - should decide on terms, “non-standard”, “custom”, etc. Just to be clear in spec.
      • Ralph explained the history of reserved vs non-reserved; reserved are pmix_ prefixed keys, it is expected that non-reserved keys are prefixed by something else (e.g., slurm_), and will use the timeout mechanism
      • There may be a use case for non-reserved keys that have the immediate timeout for keys set by the host environment (e.g., non-reserved with the reserved behavior)
  • WG Update - Client Separation / Implementation Agnostic Document Working Group
    • Dave Solt - WG reviewing get/fence/etc. to review text in standard to help with clarity, and avoid implementation specific details and instead clarity intention of behaviour in standard
  • WG Update - Slicing/Grouping of functionality Working Group
    • See pr voted in today
    • Looking at new use cases (sessions, etc.)
  • WG Update - Dynamic Workflows Working Group
    • Meeting tomorrow will give an update on effort for using Pmix in ADIOS, kubernetes, and more
  • WG Update - Storage Working Group
    • See pr voted on today
    • Will be reaching out to Adam Moody for reviving prior ideas/works discussed in early-on meetings
    • Checking data availability in advance, job-scheduler/data placement interactions
  • Open Call for New Working Group
    • Potentially tools/debuggers wg
  • Other updates/discussion
    • SPI for organization mgmt, application submitted and tentatively accepted.  SPI will be voting soon to have formal acceptance.  SPI noted that this will be the first standard body under SPI.
    • The split meeting day format helped to allow for some follow-up items. Goodness! :-)
  • Next meetings
    • Next ASC Monthly - Thursday (11-mar-2021)
    • Next ASC Quarterly - Tuesday (11-may-2021) / Thursday (13-may-2021)

Home

Working Groups

Events

Helpful How To...

ASC Quarterly Meetings

ASC Monthly Meetings

Clone this wiki locally