August 2014 - mpi-forum - mailman3.cels.anl.gov

New Agenda for September Meeting Available
by Schulz Martin 30 Aug '14

30 Aug '14

Hi all, Since I only got requests for the 2-day option and no concerns or complaints about moving the voting block, I have updated the agenda accordingly. I will go through the errata requests tomorrow and Monday (which is the deadline, in case someone still wants to add a ticket) and will add more details for the Monday afternoon agenda then. Note, that I left Wednesday on the agenda in case someone wants to use the time to meet and discuss WG (or other topics). In this case, though, please let me know, so we can give our hosts a heads-up on whether rooms are needed or not. Thanks! Martin PS: Please register ASAP! We need a complete list of people to arrange access to Riken. ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA

1 0

Reminder: Errata, Readings, and Votes must be announced by 9/1
by Schulz Martin 28 Aug '14

28 Aug '14

Hi all, Just as a reminder all errata items, readings and votes must be announced 2 weeks before the forum start in order to count. This deadline for the September meeting is next Monday 9/1. Please send all announcements to the general MPI forum list. Thanks! Martin ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA

1 0

Re: [Mpi-forum] September MPI Forum / Agenda and Reminder to Register
by Schulz Martin 28 Aug '14

28 Aug '14

Hi all, So far, I have 3 votes for option 2 and none for option 1. I take it, everyone else doesn’t care (which is fine, too)? Based on that, I would suggest to follow the “majority”. In this case, we would run from 10am to 6pm both days (Monday and Tuesday) and votes would be on Tuesday afternoon (around 4pm). Please let me know by the end of the week if there are any objections, in particular to moving the votes. Thanks! Martin ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA From: George Bosilca <bosilca(a)icl.utk.edu> Reply-To: Main mailing list <mpi-forum(a)lists.mpi-forum.org> Date: Thursday, August 28, 2014 at 7:32 AM To: Main mailing list <mpi-forum(a)lists.mpi-forum.org> Subject: Re: [Mpi-forum] September MPI Forum / Agenda and Reminder to Register Option 2, the one leaving Wednesday entirely off, would be my first vote. George. On Thu, Aug 28, 2014 at 3:53 AM, Schulz Martin <schulzm(a)llnl.gov> wrote: > Hi all, > > Based on the (sparse) feedback that I have gotten, I have assembled an > initial agenda and uploaded it to the server. Please check the agenda, > especially if you sent me a request, and double check that I didn¹t miss > anything or anybody. If something is missing, please let me know asap. > > As usual, I tried to split working groups and plenary on separate days, > with votes on day three. However, so far I have only received one WG > meeting request, which would make Monday relatively sparse. We could move > parts of Tuesday to that day and probably finish on Tuesday evening. Some > people have voiced interest in doing this. However, this would move votes > onto the same day as the second half of the plenary and I would like some > feedback on this (since the voting times are typically fixed to the last > day). So the options are: > > Monday WG time (with open, general WG time), Tuesday plenary, Wednesday > votes > > -or- > > Monday WG and plenary time (errata), Tuesday plenary and votes, Wednesday > off > > Please let me know if you have a preference one way or the other. If I > don¹t hear anything or if there are objections to moving the voting block, > we¹ll leave things as is (first option). > > Thanks! > > Martin > > PS: Please don¹t forget to register, if you haven¹t done so, by the end of > August (just a few days left). We need an exact attendee list, so people > can get access to Riken. > > ________________________________________________________________________ > Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ > CASC @ Lawrence Livermore National Laboratory, Livermore, USA > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum

1 0

Re: [Mpi-forum] September MPI Forum / Agenda and Reminder to Register
by Rolf Rabenseifner 28 Aug '14

28 Aug '14

Dear Martin, please add the following errata tickets to the agenda : #456 MPI 3.0 Errata: MPI RMA synchronization for MPI shared memory This is a large ticket that needs discussion first in the RMA-WG, second in the plenary, and hopefully some voting (corrctions and final) ____ Small Fortran corrections are: #433 MPI 3.0 Errata: MPI_WIN_ATTACH must only be for simply contiguous buffers #440 Fortran cross-reference correction (ticket 0 / errata) #441 Missing text for Fortran MPI_SUBARRAYS_SUPPORTED equals .FALSE. (errata) #442 Fortran "vector subscipts" and MPI_BOTTOM (ticket 0 / errata) #443 Fortran MPI_BOTTOM not for initialization or assignment (ticket 0 / errata) #444 Fortran and Persistent Requests (ticket 0 / errata) #445 Fortran and Permament Data Movement: "All" nonblocking (ticket 0 / errata) #446 Fortran cross-reference update (ticket-0/errata) ____ Fortran and absolute address text corrections: #431 MPI 3.0 Errata: MPI Address Errata ____ Errata and ticket-0 additions to tickets #349, #402, #404 (current status of the triple 1st vote): #421 MPI_Aint addressing arithmetic (ticket-0 additions) ____ Small errata corrections to Chap.2: #447 Chap.2 errata: Tools section is an exception: no Fortran interface #448 Chap.2 errata: Semantic term "matching" is missing #449 Chap.2 errata: Missing "in C" in wording about handle-comparison #450 Chap.2 errata: Outdated cross-reference to MPI-1 #451 Chap.2 errata: Correcting description of nonblocking #452 Chap.2 errata: Wording correction about deprecated and removed functions #453 Chap.2 errata: Avoidance of mpi_ / pmpi_ in C due to Fortran reasons #455 Chap.2 errata: Removal of duplicated text on MPI_Aint and MPI_Offset _____________________________________ The rest of this mail are only hints. ____ I also expect that Jeff may want to update, discuss and vote: #428 Add more non-profile-able MPI routines ____ I also expect that Kathryn may want to update, discuss and vote: #383 Consistency of tools information interface variable information #368 MPI_T Minor Errata ____ I also expect that Bill may want to update, discuss and vote: #438 Interpretation of MPI_WIN_BASE for shared memory windows ____ Other MPI-3.0 errata tickets that may or should be handled before MPI-3.1 is finalized: #420 Bug in neighbourhood collective example - due to missing chapter committee action, it shoud be voted in ____ The only MPI-3.0-errata ticket that may need to go MPI-4.0 is: #374 Comm_free/Type_free/Win_free: clarify attribute destructor semantics ____ Tickets #428, #368, #368, #420 may need a milestone-change to "Kobe". Otherwise, they are invisible in the Kobe tickets list: https://svn.mpi-forum.org/trac/mpi-forum-web/query?status=accepted&status=a… Best regards Rolf ----- Original Message ----- > From: "Schulz Martin" <schulzm(a)llnl.gov> > To: "Main mailing list" <mpi-forum(a)lists.mpi-forum.org> > Sent: Thursday, August 28, 2014 9:53:01 AM > Subject: [Mpi-forum] September MPI Forum / Agenda and Reminder to Register > > Hi all, > > Based on the (sparse) feedback that I have gotten, I have assembled > an > initial agenda and uploaded it to the server. Please check the > agenda, > especially if you sent me a request, and double check that I didn¹t > miss > anything or anybody. If something is missing, please let me know > asap. > > As usual, I tried to split working groups and plenary on separate > days, > with votes on day three. However, so far I have only received one WG > meeting request, which would make Monday relatively sparse. We could > move > parts of Tuesday to that day and probably finish on Tuesday evening. > Some > people have voiced interest in doing this. However, this would move > votes > onto the same day as the second half of the plenary and I would like > some > feedback on this (since the voting times are typically fixed to the > last > day). So the options are: > > Monday WG time (with open, general WG time), Tuesday plenary, > Wednesday > votes > > -or- > > Monday WG and plenary time (errata), Tuesday plenary and votes, > Wednesday > off > > Please let me know if you have a preference one way or the other. If > I > don¹t hear anything or if there are objections to moving the voting > block, > we¹ll leave things as is (first option). > > Thanks! > > Martin > > PS: Please don¹t forget to register, if you haven¹t done so, by the > end of > August (just a few days left). We need an exact attendee list, so > people > can get access to Riken. > > ________________________________________________________________________ > Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ > CASC @ Lawrence Livermore National Laboratory, Livermore, USA > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > -- Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner(a)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)

1 0

Re: [Mpi-forum] September MPI Forum / Agenda and Reminder to Register
by George Bosilca 28 Aug '14

28 Aug '14

Option 2, the one leaving Wednesday entirely off, would be my first vote. George. On Thu, Aug 28, 2014 at 3:53 AM, Schulz Martin <schulzm(a)llnl.gov> wrote: > Hi all, > > Based on the (sparse) feedback that I have gotten, I have assembled an > initial agenda and uploaded it to the server. Please check the agenda, > especially if you sent me a request, and double check that I didn¹t miss > anything or anybody. If something is missing, please let me know asap. > > As usual, I tried to split working groups and plenary on separate days, > with votes on day three. However, so far I have only received one WG > meeting request, which would make Monday relatively sparse. We could move > parts of Tuesday to that day and probably finish on Tuesday evening. Some > people have voiced interest in doing this. However, this would move votes > onto the same day as the second half of the plenary and I would like some > feedback on this (since the voting times are typically fixed to the last > day). So the options are: > > Monday WG time (with open, general WG time), Tuesday plenary, Wednesday > votes > > -or- > > Monday WG and plenary time (errata), Tuesday plenary and votes, Wednesday > off > > Please let me know if you have a preference one way or the other. If I > don¹t hear anything or if there are objections to moving the voting block, > we¹ll leave things as is (first option). > > Thanks! > > Martin > > PS: Please don¹t forget to register, if you haven¹t done so, by the end of > August (just a few days left). We need an exact attendee list, so people > can get access to Riken. > > ________________________________________________________________________ > Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ > CASC @ Lawrence Livermore National Laboratory, Livermore, USA > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >

1 0

September MPI Forum / Agenda and Reminder to Register
by Schulz Martin 28 Aug '14

28 Aug '14

Hi all, Based on the (sparse) feedback that I have gotten, I have assembled an initial agenda and uploaded it to the server. Please check the agenda, especially if you sent me a request, and double check that I didn¹t miss anything or anybody. If something is missing, please let me know asap. As usual, I tried to split working groups and plenary on separate days, with votes on day three. However, so far I have only received one WG meeting request, which would make Monday relatively sparse. We could move parts of Tuesday to that day and probably finish on Tuesday evening. Some people have voiced interest in doing this. However, this would move votes onto the same day as the second half of the plenary and I would like some feedback on this (since the voting times are typically fixed to the last day). So the options are: Monday WG time (with open, general WG time), Tuesday plenary, Wednesday votes -or- Monday WG and plenary time (errata), Tuesday plenary and votes, Wednesday off Please let me know if you have a preference one way or the other. If I don¹t hear anything or if there are objections to moving the voting block, we¹ll leave things as is (first option). Thanks! Martin PS: Please don¹t forget to register, if you haven¹t done so, by the end of August (just a few days left). We need an exact attendee list, so people can get access to Riken. ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA

1 0

fftw and linear solver
by LikunTan 14 Aug '14

14 Aug '14

Hi all, I have a general question of implementing FFTW and KSPSolve in Petsc. My code consists mainly two parts: part1: calculating the values of arrays, which takes a convolution form. So I am using FFTW for high efficiency. part2: solve a linear system Ax=b, where A is constant and b is provided from part1. In part1, the array size is 512*512, and I found that taking np=8 gives optimal time cost ~0.025s. But I have 225 of such arrays and they can be done speparately. In part2, the vector b is composed of the arrays calculated before, the size of b is about 50000. It takes around 1s to do KSPSolve when choosing np=8, but futher increase of np will slow down KSPSolve by a factor of 3~4. To save time, I use MPI_Comm_split to divide MPI_COMM_WORLD into 15 sub-communicators and do FFTW in parallel. Due to the overhead in transferring data and increasing time cost in part2, I found by taking np=16, and using one processor to solve each FFTW (and solve 15 FFTW in parallel) is the fastest. The total time for part1 and part2 is about 3s. I have tried my best to get the optimal result, but I am not sure if there is any way to improve efficiency. I would very much appreciate your comments and suggestions. Thanks, Likun

1 0

Countdown to MPI Forum Meeting / September 2014 in Kobe
by Martin Schulz 08 Aug '14

08 Aug '14

Hi all, We are getting closer to the next MPI forum meeting and I want to make sure that everyone is aware of the deadlines and tasks for it: - If you are attending, please register for the forum before the end of the month - this is necessary this time to enable access to the site. The registration link is, as always, on the logistics page. - If any working group wants to bring a ticket forward for reading and/or vote, you must do so by September 1 (two weeks ahead of the forum) - note this is labor day, so please plan ahead. Also, please let me know informally as soon as possible so I can plan the agenda. - I already pinged all working group chairs, but as a general reminder - please send me any agenda items as soon as possible so I can have a reasonable agenda together. So far I have only received very few items. Note that based on the timeline we decided in the previous meetings, this is the last MPI forum meeting to have any errata to be included in MPI 3.1 - after this meeting we¹ll only focus on pulling the document together. We¹ll talk about the process for this at the meeting. Thanks! Martin PS: If someone is still having problems finding a hotel in Kobe, please let me know. ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://scalability.llnl.gov/ CASC @ Lawrence Livermore National Laboratory, Livermore, USA

1 0

Registration for the Sept meeting
by Richard Graham 03 Aug '14

03 Aug '14

The registration for the Sept meeting is up and available at: http://a3.acteva.com/orderbooking/bookEvent/A340722 Rich

1 0