[Mpi-forum] Process towards MPI 4.0 / Final Steps

Tue Mar 30 11:51:45 CDT 2021

Dear Wesley,

To not saying that such while loops with MPI_Request_get_status are correct,
means that the application codes doing so are broken.
With adding a new function and explicitly staying again 
(same as for MPI_Win_test 20 years ago) that such while loops are 
allowed, gives now a clear hint that MPI_Request_get_status
does not allow such programming.
We as the MPI forum decided on Feb. 2021 that we do not want to
give this inconsistent impression. Therefore we decided to put this
into the MPI-4.0 set of late corrections.
It is well understood that it was overseen since MPI-2
and the implementors already implemented it without informing the forum.
The fix was therefore prepared during the meeting, read and set up for
the RC vote. RC vote was done. Therefore it has to go to the FRM vote.

It was okay to ask again on last Wednesday, whether this voting process should be
continued and my clear answer based on the reasons above is a clear yes.

One major duty of the read-whole-standard-review-process is, to find such
inconsistencies and correct them.
And as we can see, the process worked fine.

If you want again to discuss it tomorrow, that's okay.
If you ask for an additional reviewer, because you may expect 
the whole forum has overseen an important technical aspect, that's also okay.

Best regards
Rolf

PS: With the other issue 492, your proposal to move to MPI-4.1 is agreed.

----- Original Message -----
> From: "Wesley Bland" <work at wesbland.com>
> To: "Rolf Rabenseifner" <rabenseifner at hlrs.de>, "Main MPI Forum mailing list" <mpi-forum at lists.mpi-forum.org>
> Sent: Tuesday, March 30, 2021 5:51:22 PM
> Subject: Re: [Mpi-forum] Process towards MPI 4.0 / Final Steps

> This was removed from 4.0 because it was not considered an “emergency fix”.
> That’s not to say that it’s not worthwhile and we can’t or shouldn’t do it for
> 4.1, just that we’re going to remove all changes from the list that are not
> required so we can make sure to finish 4.0 in June.
> 
> This can still be read in June and voted in to 4.1 in the usual time frame after
> that (1st in September, 2nd in December).
> 
> Am I misunderstanding that there’s something breaking by not having this in 4.0?
> 
> Thanks,
> Wes
> 
>> On Mar 30, 2021, at 4:03 AM, Rolf Rabenseifner via mpi-forum
>> <mpi-forum at lists.mpi-forum.org> wrote:
>> 
>> Dear Martin, Wesley, Bill and all,
>> 
>>> Please let us know if you have any questions or concerns - we will also address
>>> this in the meeting next week.
>> 
>> I detected that also Issue (K=) 468 was moved to MPI-4.1.
>> 
>> I updated the issues's abstract with a summary of the discussion
>> (which was partially within the Parrived progress discussion)
>> at the Feb. 2021 meeting.
>> 
>> Adding a 4th such test routine (after MPI_Test, Request_get_status, Win_test,
>> Parrived),
>> it gets significantly more evident if for only one routine, Request_get_status,
>> there is no such written progress guarantee in the standard.
>> 
>> Therefore, we did this analysis and reading of adding at the Feb. meeting
>> for voting it into the standard in June.
>> 
>> There is no reason to break this result from the Feb. 2021 meeting.
>> 
>> Therefore, I put https://github.com/mpi-forum/mpi-issues/issues/468
>> back to the triage.
>> 
>> @Martin please put it for tomorrow on the agenda.
>> 
>> Best regards
>> Rolf
>> 
>> 
>> ----- Original Message -----
>>> From: "Main MPI Forum mailing list" <mpi-forum at lists.mpi-forum.org>
>>> To: "Main MPI Forum mailing list" <mpi-forum at lists.mpi-forum.org>
>>> Cc: "Martin Schulz" <schulzm at in.tum.de>
>>> Sent: Wednesday, March 24, 2021 7:55:11 PM
>>> Subject: [Mpi-forum] Process towards MPI 4.0 / Final Steps
>> 
>>> Hi all,
>>> 
>>> For those of you who had to miss the virtual MPI Forum meeting today, we had a
>>> discussion on the updated process towards MPI 4.0 and how to handle the final
>>> tickets.
>>> 
>>> The recording is at (with the right time-marker to this discussion):
>>> https://youtu.be/h9MO-BmbfdQ?t=2125
>>> 
>>> The slides are at:
>>> https://github.com/mpi-forum/mpi-forum.github.io/raw/master/slides/2021/03/2021-03-24-rcm-frm-procedures.pptx
>>> 
>>> The quick summary is that we will limit the tickets that are still considered
>>> for MPI 4.0 to only emergency fixes that avoid breaking the standard and that
>>> are directly related to new items/procedures/concepts/text added specifically
>>> for MPI 4.0. Otherwise we fear (and we are seeing this already) that we may get
>>> clogged up with more and more changes that will not get the needed vetting and
>>> - equally important - that we run the risk of some issues not being accepted
>>> during the vote, which means the FRM is not the final document, which is
>>> problematic.
>>> 
>>> Wesley will be going through the board and classifying all issues based on that
>>> (see also slide 11) and then move tickets that do not meet the criteria to MPI
>>> 4.1 (changes that are already in, will stay , though). For this, we agreed to
>>> temporarily assume the role of a release manager that will advance tickets once
>>> they are ready (and that includes sufficient reviews by the CCs) and meet the
>>> criteria.
>>> 
>>> Note, though, this does not mean that the items tagged as 4.1 are postponed
>>> indefinitely - the plan is to start working on them right after MPI 4.0, if
>>> time permits, already in the June meeting. We should strive for getting a
>>> cleaned-up MPI 4.1 out the door soon, preferably before we add (m)any new
>>> items, but that part we will also discuss in the June meeting.
>>> 
>>> Please let us know if you have any questions or concerns - we will also address
>>> this in the meeting next week.
>>> 
>>> Thanks!
>>> 
>>> Martin
>>> (also in the name of Bill and Wesley)
>>> 
>>> 
>>> 
>>> --
>>> 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
>>> 
>>> 
>>> 
>>> 
>>> _______________________________________________
>>> mpi-forum mailing list
>>> mpi-forum at lists.mpi-forum.org
>>> https://lists.mpi-forum.org/mailman/listinfo/mpi-forum
>> 
>> --
>> Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner at hlrs.de .
>> High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 .
>> University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 .
>> Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner .
>> Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) .
>> _______________________________________________
>> mpi-forum mailing list
>> mpi-forum at lists.mpi-forum.org
> > https://lists.mpi-forum.org/mailman/listinfo/mpi-forum

-- 
Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner at hlrs.de .
High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 .
University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 .
Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner .
Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) .