May 2016 - mpi-forum - mailman3.cels.anl.gov

MPI Forum Registration Reminder
by Brian Smith 31 May '16

31 May '16

Hi everyone, As a reminder, registration is required for the MPI Forum next week so we can pass the information along to Microsoft and get everyone appropriate access. So far we only have 9 registrations. The registration link is on the logistics page for the meeting. Thanks. Brian Smith brians(a)mellanox.com<mailto:[email protected]>

1 0

Re: [Mpi-forum] Next MPI Forum Locations
by Jeff Hammond 26 May '16

26 May '16

> > > As an additional item to think about: in 2018, EuroMPI will again be > outside the US and so we can go back to our normal routine. Additionally, > though, SIAM-PP will be in Tokyo in the spring of 2018 and I was approached > by the organizers whether we could co-locate the March forum meeting. The > idea was to help each other increase attendance based on the observation > that a good number of the regular MPI forum attendees were also part of > SIAM-PP 2016, which was in Paris. Are we interested in pursuing this? If > so, in addition or instead of EuroMPI? > > I like SIAM-PP instead of EuroMPI because I am far more likely to attend the former, and I shared the observation about how many folks were in Paris for PP16. However, it will likely hurt EuroMPI, because some institutions' travel policies may not let folks attend foreign conferences that are not top-tier, whereas attending an industry working group of which the institution is a voting member is a compelling reason to travel. It might be good to poll people on whether moving the MPI Forum decreases the likelihood that they or their colleagues will submit to EuroMPI. Jeff -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/

1 0

Next MPI Forum Locations
by Schulz Martin 26 May '16

26 May '16

Hi all, During the upcoming MPI forum meeting in June we need to plan for the next set of locations and dates in 2017. The goal will, as always, be to a) find locations that are cost effective for the forum; b) times that don¹t class with other HPC events; and c) to ensure the most people can come. For next year, though, we have a few anomalies that need to be discussed and so I am writing this email as a heads-up. If you want to respond with some early comments, please do so. Also, if you can¹t make it to the meeting in June, but would like your voice heard, please send me email or respond to the group. We¹ll take all of this into account during the discussion and decision in Bellevue. The biggest anomaly is that EuroMPI is in Chicago next year. The exact dates are: September 25-28, 2017 Since EuroMPI was traditionally our 1 in 4 slot to be not in the US, we have to decide what do do as I see it, we have three options: * Not do any meeting outside the US next year this will severely impact the ability to attend for some folks, especially from Europe, as far as I understand. * Pick one of the other meetings and go outside the US for that instead of the September meeting without EuroMPI, this can severely impact US-based attendance. The suggestion in this case would be to combine the forum meeting with a large event that many may be traveling to anyway ISC in Frankfurt (6/17-6/21) could fit that bill and RWTH Aachen in conjunction with JSC has tentative agreed to explore hosting us in Aachen the week before ISC (only about a 1-2 hour train ride between Aachen and Frankfurt and the facilities in Aachen have worked well for previous similar meetings) - thanks to Marc-Andre for helping explore this option! * We could stick with the schedule and go outside the US in September anyway (without overlapping with EuroMPI, of course :) ). If we hold the MPI forum meeting next September in conjunction with EuroMPI in Chicago, we have the additional problem that we likely need to leave a gap of a full weekend between the forum and the conference (based on space availability and having to avoid weekends) or pay extra. Is a co-location still useful in this case? Aside from these issues, we¹ll have the usual discussion of good locations we had previously been hosted by Cisco in San Jose and various Microsoft locations (Chicago, Portland, and now Bellevue) as well as we¹ll be in Dallas at an IBM site in December. Would anyone of these hosts be willing to host the forum again? SGI has also offered to host us in Eagan next year (thanks to all previous and current hosts!). Anyone else willing to host the forum in the US? As an additional item to think about: in 2018, EuroMPI will again be outside the US and so we can go back to our normal routine. Additionally, though, SIAM-PP will be in Tokyo in the spring of 2018 and I was approached by the organizers whether we could co-locate the March forum meeting. The idea was to help each other increase attendance based on the observation that a good number of the regular MPI forum attendees were also part of SIAM-PP 2016, which was in Paris. Are we interested in pursuing this? If so, in addition or instead of EuroMPI? Comments, Thoughts? Thanks! Martin ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA

1 0

Re: [Mpi-forum] Draft agenda posted for June 2016 meeting
by Schulz Martin 23 May '16

23 May '16

Hi all, I got a few last minute updates the latest agenda draft is online. Please check if anything is incorrect. Thanks! Martin ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA From: Schulz Martin <schulzm(a)llnl.gov> Date: Monday, May 23, 2016 at 9:06 PM To: Main mailing list <mpi-forum(a)lists.mpi-forum.org> Subject: Draft agenda posted for June 2016 meeting Hi all, Based on the input I have gotten so far, I have posted a draft agenda for the June meeting at: http://meetings.mpi-forum.org/secretary/2016/06/agenda.php If you see any mistake or if I missed any request, please let me know as soon as possible. Below is a full list of all official readings and votes lot¹s of homework for everyone this time :) Thanks! Martin Reading#28 <https://github.com/mpi-forum/mpi-issues/issues/28> : Add Catastrophic and Noncatastrophic Error CodesWesley #1 <https://github.com/mpi-forum/mpi-issues/issues/1> : Clarify MPI_ERRORS_ARE_FATAL scope of abortWesley #3 <https://github.com/mpi-forum/mpi-issues/issues/3> : Define new MPI Error handler for subcommunicator abortWesley #20 <https://github.com/mpi-forum/mpi-issues/issues/20> : ULFM part 1Aurelien #21 <https://github.com/mpi-forum/mpi-issues/issues/21> : ULFM part 2Aurelien #22 <https://github.com/mpi-forum/mpi-issues/issues/22> : ULFM part 3Aurelien Procedures Reading#35 <https://github.com/mpi-forum/mpi-issues/issues/35> : Procedure Changes: Allow move of voting block as an exceptionMartin Change Reading#26 <https://github.com/mpi-forum/mpi-issues/issues/26> : Deprecate point-to-point send cancelDan Errata Reading#52 <https://github.com/mpi-forum/mpi-issues/issues/52> : Erratum to remove info key propagation from MPI_COMM_DUP (ticketErrta)Dan OtherPersistence WGTony Reading#53 <https://github.com/mpi-forum/mpi-issues/issues/53> : MPI_COMM_DUP_WITH_INFO can be used to explicitly propagate info hints (ticketIDWI)Dan #11 <https://github.com/mpi-forum/mpi-issues/issues/11> : Adds communicator info keys and allows info assertions (ticketInfo)Dan #6 <https://github.com/mpi-forum/mpi-issues/issues/6> : MPIR: Include line numbers in specificationKathryn ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA

1 0

Reading: rule change about ballot times
by Jeff Squyres (jsquyres) 23 May '16

23 May '16

Per discussion at the last Forum meeting, Martin and I would like to read the proposed rules change at the Bellevue meeting: https://github.com/mpi-forum/mpi-issues/issues/35 Here's the old text: The dates/times for official ballots will not change after two weeks prior to the beginning of the meeting to allow attendees to schedule their travel appropriately. Here's the new text: The dates/times for official ballots will allow attendees to schedule their travel appropriately, except upon unanimous agreement of all OOE organizations who have at least one individual registered for this physical MPI Forum meeting before the meeting started. Upon such agreement, official ballots can only be moved to a different date/time within the same physical MPI Forum meeting, provided that the number of these organizations satisfies the meeting quorum requirements. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Deprecate MPI_SIZEOF in MPI-4.0
by Jeff Squyres (jsquyres) 23 May '16

23 May '16

The Fortran WG would like to read this PR at the Bellevue meeting: https://github.com/mpi-forum/mpi-issues/issues/51 -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Draft agenda posted for June 2016 meeting
by Schulz Martin 23 May '16

23 May '16

Hi all, Based on the input I have gotten so far, I have posted a draft agenda for the June meeting at: http://meetings.mpi-forum.org/secretary/2016/06/agenda.php If you see any mistake or if I missed any request, please let me know as soon as possible. Below is a full list of all official readings and votes lot¹s of homework for everyone this time :) Thanks! Martin Reading#28 <https://github.com/mpi-forum/mpi-issues/issues/28> : Add Catastrophic and Noncatastrophic Error CodesWesley #1 <https://github.com/mpi-forum/mpi-issues/issues/1> : Clarify MPI_ERRORS_ARE_FATAL scope of abortWesley #3 <https://github.com/mpi-forum/mpi-issues/issues/3> : Define new MPI Error handler for subcommunicator abortWesley #20 <https://github.com/mpi-forum/mpi-issues/issues/20> : ULFM part 1Aurelien #21 <https://github.com/mpi-forum/mpi-issues/issues/21> : ULFM part 2Aurelien #22 <https://github.com/mpi-forum/mpi-issues/issues/22> : ULFM part 3Aurelien Procedures Reading#35 <https://github.com/mpi-forum/mpi-issues/issues/35> : Procedure Changes: Allow move of voting block as an exceptionMartin Change Reading#26 <https://github.com/mpi-forum/mpi-issues/issues/26> : Deprecate point-to-point send cancelDan Errata Reading#52 <https://github.com/mpi-forum/mpi-issues/issues/52> : Erratum to remove info key propagation from MPI_COMM_DUP (ticketErrta)Dan OtherPersistence WGTony Reading#53 <https://github.com/mpi-forum/mpi-issues/issues/53> : MPI_COMM_DUP_WITH_INFO can be used to explicitly propagate info hints (ticketIDWI)Dan #11 <https://github.com/mpi-forum/mpi-issues/issues/11> : Adds communicator info keys and allows info assertions (ticketInfo)Dan #6 <https://github.com/mpi-forum/mpi-issues/issues/6> : MPIR: Include line numbers in specificationKathryn ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA

1 0

Re: [Mpi-forum] link to registration page for June 2016 meeting?
by Schulz Martin 23 May '16

23 May '16

Thanks, Brian! I have added the link to the page this should appear shortly on the public server. Martin ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA From: mpi-forum <mpi-forum-bounces(a)mpi-forum.org> on behalf of Brian Smith <brians(a)mellanox.com> Reply-To: Main MPI Forum mailing list <mpi-forum(a)mpi-forum.org> Date: Monday, May 23, 2016 at 6:38 PM To: Main MPI Forum mailing list <mpi-forum(a)mpi-forum.org>, Main mailing list <mpi-forum(a)lists.mpi-forum.org> Subject: Re: [Mpi-forum] link to registration page for June 2016 meeting? The meeting registration link is: http://www.eventbrite.com/e/mpi-forum-seattle-tickets-25514658029 <http://www.eventbrite.com/e/mpi-forum-seattle-tickets-25514658029?aff=utm_s ource%3Deb_email%26utm_medium%3Demail%26utm_campaign%3Dnew_event_email&utm_t erm=eventurl_text> I thought I had permission to update meetings.mpi-forum.org but no luck so far. The link will hopefully be there shortly though. Also How much need is there to try to get a block of hotel rooms near the Microsoft Office? We haven¹t had much luck getting a block without an actual commitment. Thanks. Brian Smith brians(a)mellanox.com From: mpi-forum [mailto:[email protected]] On Behalf Of Pritchard Jr., Howard Sent: Monday, May 23, 2016 1:06 PM To: mpi-forum(a)lists.mpi-forum.org Subject: [Mpi-forum] link to registration page for June 2016 meeting? HI Folks, Anyone know when the link to the registration page for the upcoming forum meeting will be available? Howard -- Howard Pritchard HPC-DES Los Alamos National Laboratory

1 0

Re: [Mpi-forum] link to registration page for June 2016 meeting?
by Brian Smith 23 May '16

23 May '16

The meeting registration link is: http://www.eventbrite.com/e/mpi-forum-seattle-tickets-25514658029<http://www.eventbrite.com/e/mpi-forum-seattle-tickets-25514658029?aff=utm_s…> I thought I had permission to update meetings.mpi-forum.org but no luck so far. The link will hopefully be there shortly though. Also - How much need is there to try to get a block of hotel rooms near the Microsoft Office? We haven't had much luck getting a block without an actual commitment. Thanks. Brian Smith brians(a)mellanox.com<mailto:[email protected]> From: mpi-forum [mailto:[email protected]] On Behalf Of Pritchard Jr., Howard Sent: Monday, May 23, 2016 1:06 PM To: mpi-forum(a)lists.mpi-forum.org Subject: [Mpi-forum] link to registration page for June 2016 meeting? HI Folks, Anyone know when the link to the registration page for the upcoming forum meeting will be available? Howard -- Howard Pritchard HPC-DES Los Alamos National Laboratory

1 0

Communicator Info Reading for June 2016 Meeting
by Jim Dinan 23 May '16

23 May '16

Hi All, The P2P working group has revised the communicator info proposal and split it into three parts based on feedback at the last meeting: - #52 - Erratum to remove info key propagation from MPI_COMM_DUP (ticketErrta) - #53 - MPI_COMM_DUP_WITH_INFO can be used to explicitly propagate info hints (ticketIDWI) - #11 - Adds communicator info keys and allows info assertions (ticketInfo) We would like to read each proposal at the June meeting. Note that #52 is an erratum and the others are full proposals. This breakdown was arrived at based on the feedback from the last meeting. All three proposals are still contained in the same document and pull request (attached, and referenced from #11) with the "ticket" names shown above. Dan Holmes will conduct the readings. Thanks, ~Jim.

1 0