[mpiwg-persistence] Unused persistent collectives info keys in Annex

HOLMES Daniel d.holmes at epcc.ed.ac.uk
Fri Aug 30 03:33:40 CDT 2019


Hi Wesley,

In which case, the new info key that we want in connection with persistent collectives (#83) is either an errata for an as-yet-unreleased version of the MPI Standard (except, perhaps, it could fix the SC18Draft version, which was kinda published) OR it has already missed the MPI-4.0 deadline.

I would prefer that we follow the errata procedure (reading in Dec, only-vote in 2020Q1) and write the change-log for it as a tweak to the existing MPI-4.0 change-log entry for persistent collectives - as if the info key ticket had always been part of the main persistent collectives ticket.

Note that removing the spurious info keys in the Annex could be a ticket-0/chapter committee tidy-up of the chapter, which can be done any time without formal approval by the Forum at large.

Cheers,
Dan.
—
Dr Daniel Holmes PhD
Architect (HPC Research)
d.holmes at epcc.ed.ac.uk<mailto:d.holmes at epcc.ed.ac.uk>
Phone: +44 (0) 131 651 3465
Mobile: +44 (0) 7940 524 088
Address: Room 2.09, Bayes Centre, 47 Potterrow, Central Area, Edinburgh, EH8 9BT
—
The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.
—

On 29 Aug 2019, at 21:10, Wesley Bland <work at wesbland.com<mailto:work at wesbland.com>> wrote:

According to our plan, this was the last meeting at which new proposals could be read in order to make that deadline. I’d be very hesitant to add yet another proposal type beyond our current three (normal, errata, no-no) in order to rush things in.

On Aug 29, 2019, at 9:01 AM, HOLMES Daniel via mpiwg-persistence <mpiwg-persistence at lists.mpi-forum.org<mailto:mpiwg-persistence at lists.mpi-forum.org>> wrote:

Hi Martin,

Procedurally, do we have enough time (that is, enough face-to-face meetings) to read a full proposal and get both the 1st vote and the 2nd vote done before the deadline imposed by our intention to release MPI-4.0 at SC20 which requires prior ratification of a release candidate? Note that #83 is not on the agenda for Zurich.

The errata procedure only requires one formal vote, which shaves off one face-to-face meeting from the critical path (vs. the full proposal procedure) - although I agree it is a little unclear which version is being corrected. Using errata for new functionality suggests misuse of the “correct a small mistake” procedure as a “short-circuit for a small addition” procedure.

OTOH, maybe we should innovate and approve such a “small addition” procedure? Maybe some changes need two formal votes but others only need one? Could we introduce a no-no-vote type of approval-by-acclamation at the end of the formal reading meeting to decide whether the proposal just read is “small” (needs one vote) or not (needs two votes)?

Alternatively, perhaps the rules could permit 2nd votes at the start of the release candidate meeting, with the thing-to-be-gone-through-page-by-page being affected by the result(s) of that voting block with a short intermission while the document editor merges the successful pull requests (or a longer intermission if there are any merge conflicts that need fixing)?

Cheers,
Dan.
—
Dr Daniel Holmes PhD
Architect (HPC Research)
d.holmes at epcc.ed.ac.uk<mailto:d.holmes at epcc.ed.ac.uk>
Phone: +44 (0) 131 651 3465
Mobile: +44 (0) 7940 524 088
Address: Room 2.09, Bayes Centre, 47 Potterrow, Central Area, Edinburgh, EH8 9BT
—
The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.
—

On 29 Aug 2019, at 14:11, Martin Schulz via mpiwg-persistence <mpiwg-persistence at lists.mpi-forum.org<mailto:mpiwg-persistence at lists.mpi-forum.org>> wrote:

Hi Tony, all,

I hope I fully followed this :)

Doing an errata item seems a bit backwards to me, as we haven’t published the standard, yet. Also, removing it and then re-adding it with 83 seems odd. What about just removing the two keys that are not needed with 83 through the no/no process? Wouldn’t that be all that is needed?

Martin


—
Prof. Dr. Martin Schulz, Chair of Computer Architecture and Parallel Systems
Department of Informatics, TU-Munich, Boltzmannstraße 3, D-85748 Garching
Member of the Board of Directors at the Leibniz Supercomputing Centre (LRZ)
Email: schulzm at in.tum.de<mailto:schulzm at in.tum.de>








On 29. Aug 2019, at 12:31, Skjellum, Anthony <Tony-Skjellum at utc.edu<mailto:Tony-Skjellum at utc.edu>> wrote:

Hi, thank you...  we will issue an appropriate change to remove these... they are not part of ticket #25 as approved...and so we must issue an erratum on ticket #25.  They split off into ticket #83 in March, 2018.

We itrerated multiple times on having or not having these info keys as part of ticket #25 and we must have missed removing completely.

Such info keys  are part of a second viable ticket we that we split out are still working on ... none belong in the standard yet.  The first key value is part of ticket 83 as of now...
which  will get a no no plus vote  in December (it had its reading priorly), introducing the first key... since acceptance is anticipated.  Only examples for this ticket  need to be fixed. Plus now evidently Annex 1.5 :-) too.

The second and third key values are currently part of no proposed ticket.
They are vestigial.

So, our plan is to remove all mention of them with appropriate process and then watch for the passage of 83. :-)

I will make sure that no no vote ticket for 83 also includes adding the first key to the Annex 1.5.

I will ask Bill Gropp and Martin Schulz  for advice on fixing this.  They are cc’d.

Tony

Anthony Skjellum, PhD
205-807-4968


On Aug 29, 2019, at 5:12 AM, Joseph Schuchart <schuchart at hlrs.de<mailto:schuchart at hlrs.de>> wrote:

Tony,

I just rebased my MPI PR to the MPI 4.x branch. Looking at the info keys, I found that there are info keys in Annex A.1.5 that stem from the persistent collectives chapter but the text mentioning them was commented out there (git blame mentions you as the last author of that part). It seems that the info keys in the annex are a relic of this text. The keys in question are:

mpi_assert_strict_start_ordering
mpi_optimization_goal
mpi_reuse_count

Are you planning to revive these keys or should they be removed?

Best regards,
Joseph
--
Dipl.-Inf. Joseph Schuchart
High Performance Computing Center Stuttgart (HLRS)
Nobelstr. 19
D-70569 Stuttgart

Tel.: +49(0)711-68565890
Fax: +49(0)711-6856832
E-Mail: schuchart at hlrs.de<mailto:schuchart at hlrs.de>

_______________________________________________
mpiwg-persistence mailing list
mpiwg-persistence at lists.mpi-forum.org<mailto:mpiwg-persistence at lists.mpi-forum.org>
https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence

_______________________________________________
mpiwg-persistence mailing list
mpiwg-persistence at lists.mpi-forum.org<mailto:mpiwg-persistence at lists.mpi-forum.org>
https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-persistence/attachments/20190830/46ef523e/attachment-0001.html>


More information about the mpiwg-persistence mailing list