August 2013 - mpi-forum - mailman3.cels.anl.gov

Re: [Mpi-forum] MPI "Allocate receive" proposal
by David Goodell (dgoodell) 26 Aug '13

26 Aug '13

I don't think that you currently have any guarantee about what happens if MPI_Alloc_mem is unable to allocate memory. See MPI-3.0, p.340, l.16 (emphasis mine): """ The function MPI_ALLOC_MEM *may* return an error code of class MPI_ERR_NO_MEM to indicate it failed because memory is exhausted. """ Jeff and I don't have any text right now, but we could easily include a similarly weak statement. -Dave On Aug 26, 2013, at 10:27 AM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > > Can you clarify a bit more? > > Is that a fatal failure at that point? Or is it more like MPI_Alloc_mem failing where I can retry later or just do a regular malloc (in this case, a regular MPI_RECV). > > -- Pavan > > On 08/26/2013 10:19 AM, David Goodell (dgoodell) wrote: >> I'd say that's implementation defined behavior, no different than hitting any other system resource limitation right now. >> >> -Dave >> >> On Aug 26, 2013, at 10:14 AM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: >> >>> Jeff, >>> >>> This should probably be moved to the point-to-point WG, but I'll comment on this thread for now. >>> >>> Such a proposal was brought up at the very early stage of the MPI-2.1 process (I don't remember by whom). The main concern at that point was what happens when the memory allocation fails. Will the receive fail? What happens to the sent message? >>> >>> Can you work these details into your slides? >>> >>> -- Pavan >>> >>> On 08/26/2013 10:10 AM, Jeff Squyres (jsquyres) wrote: >>>> Dave Goodell and I have a proposal that we've socialized a bit around with other Forum members, and would now like larger Forum feedback. I'll be presenting the attached slides on the concept of an "allocate receive" in Madrid (3:30-4pm on Thursday). >>>> >>>> There's no text or ticket yet; this is an idea that we want to get feedback on before working up a full proposal. >>>> >>>> >>>> >>>> _______________________________________________ >>>> mpi-forum mailing list >>>> mpi-forum(a)lists.mpi-forum.org >>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >>>> >>> >>> -- >>> Pavan Balaji >>> http://www.mcs.anl.gov/~balaji >>> _______________________________________________ >>> mpi-forum mailing list >>> mpi-forum(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >> >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >> > > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji

1 0

Re: [Mpi-forum] [EXTERNAL] Re: MPI "Allocate receive" proposal
by Barrett, Brian W 26 Aug '13

26 Aug '13

On 8/26/13 9:21 AM, "Jeff Squyres (jsquyres)" <jsquyres(a)cisco.com> wrote: >On Aug 26, 2013, at 11:14 AM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > >> This should probably be moved to the point-to-point WG, but I'll >>comment on this thread for now. > >Is that WG active? If so, we'd be happy to move it to there. The working group mailing list is active and people read them. I think we've struggled for a proposal that causes enough interest that we can escape our current (inert) state. Brian -- Brian W. Barrett Scalable System Software Group Sandia National Laboratories

1 0

Re: [Mpi-forum] MPI "Allocate receive" proposal
by Jeff Hammond 26 Aug '13

26 Aug '13

If errors are fatal, failure to allocate is fatal. It is as if MPI_Alloc_mem fails. If not fatal, then Arecv returns error code of appropriate nature. This shouldn't be treated as resource exhaustion IMO. Jeff Sent from my iPhone On Aug 26, 2013, at 10:14 AM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > Jeff, > > This should probably be moved to the point-to-point WG, but I'll comment on this thread for now. > > Such a proposal was brought up at the very early stage of the MPI-2.1 process (I don't remember by whom). The main concern at that point was what happens when the memory allocation fails. Will the receive fail? What happens to the sent message? > > Can you work these details into your slides? > > -- Pavan > > On 08/26/2013 10:10 AM, Jeff Squyres (jsquyres) wrote: >> Dave Goodell and I have a proposal that we've socialized a bit around with other Forum members, and would now like larger Forum feedback. I'll be presenting the attached slides on the concept of an "allocate receive" in Madrid (3:30-4pm on Thursday). >> >> There's no text or ticket yet; this is an idea that we want to get feedback on before working up a full proposal. >> >> >> >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji > _______________________________________________ > 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] MPI "Allocate receive" proposal
by Pavan Balaji 26 Aug '13

26 Aug '13

Can you clarify a bit more? Is that a fatal failure at that point? Or is it more like MPI_Alloc_mem failing where I can retry later or just do a regular malloc (in this case, a regular MPI_RECV). -- Pavan On 08/26/2013 10:19 AM, David Goodell (dgoodell) wrote: > I'd say that's implementation defined behavior, no different than hitting any other system resource limitation right now. > > -Dave > > On Aug 26, 2013, at 10:14 AM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > >> Jeff, >> >> This should probably be moved to the point-to-point WG, but I'll comment on this thread for now. >> >> Such a proposal was brought up at the very early stage of the MPI-2.1 process (I don't remember by whom). The main concern at that point was what happens when the memory allocation fails. Will the receive fail? What happens to the sent message? >> >> Can you work these details into your slides? >> >> -- Pavan >> >> On 08/26/2013 10:10 AM, Jeff Squyres (jsquyres) wrote: >>> Dave Goodell and I have a proposal that we've socialized a bit around with other Forum members, and would now like larger Forum feedback. I'll be presenting the attached slides on the concept of an "allocate receive" in Madrid (3:30-4pm on Thursday). >>> >>> There's no text or ticket yet; this is an idea that we want to get feedback on before working up a full proposal. >>> >>> >>> >>> _______________________________________________ >>> mpi-forum mailing list >>> mpi-forum(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >>> >> >> -- >> Pavan Balaji >> http://www.mcs.anl.gov/~balaji >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Re: [Mpi-forum] MPI "Allocate receive" proposal
by Rusty Lusk 26 Aug '13

26 Aug '13

It's not quite the same. If your buffer allocation (via malloc) after a probe fails, you can decide what to do (e.g. free some other memory and try again.) The message isn't lost, and you may be in an unhappy state but not an undefined state. I like the idea except for the introduction of the undefined result. Perhaps this could be fixed. Rusty On Monday,Aug 26, 2013, at 10:21 AM, Jeff Squyres (jsquyres) wrote: > On Aug 26, 2013, at 11:14 AM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > >> This should probably be moved to the point-to-point WG, but I'll comment on this thread for now. > > Is that WG active? If so, we'd be happy to move it to there. > >> Such a proposal was brought up at the very early stage of the MPI-2.1 process (I don't remember by whom). The main concern at that point was what happens when the memory allocation fails. Will the receive fail? What happens to the sent message? > > I think Dave answered that well: it's just like any other resource exhaustion issue with a plain vanilla MPI receive. > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/ > > _______________________________________________ > 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] MPI "Allocate receive" proposal
by Jeff Squyres (jsquyres) 26 Aug '13

26 Aug '13

On Aug 26, 2013, at 11:14 AM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > This should probably be moved to the point-to-point WG, but I'll comment on this thread for now. Is that WG active? If so, we'd be happy to move it to there. > Such a proposal was brought up at the very early stage of the MPI-2.1 process (I don't remember by whom). The main concern at that point was what happens when the memory allocation fails. Will the receive fail? What happens to the sent message? I think Dave answered that well: it's just like any other resource exhaustion issue with a plain vanilla MPI receive. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Re: [Mpi-forum] MPI "Allocate receive" proposal
by David Goodell (dgoodell) 26 Aug '13

26 Aug '13

I'd say that's implementation defined behavior, no different than hitting any other system resource limitation right now. -Dave On Aug 26, 2013, at 10:14 AM, Pavan Balaji <balaji(a)mcs.anl.gov> wrote: > Jeff, > > This should probably be moved to the point-to-point WG, but I'll comment on this thread for now. > > Such a proposal was brought up at the very early stage of the MPI-2.1 process (I don't remember by whom). The main concern at that point was what happens when the memory allocation fails. Will the receive fail? What happens to the sent message? > > Can you work these details into your slides? > > -- Pavan > > On 08/26/2013 10:10 AM, Jeff Squyres (jsquyres) wrote: >> Dave Goodell and I have a proposal that we've socialized a bit around with other Forum members, and would now like larger Forum feedback. I'll be presenting the attached slides on the concept of an "allocate receive" in Madrid (3:30-4pm on Thursday). >> >> There's no text or ticket yet; this is an idea that we want to get feedback on before working up a full proposal. >> >> >> >> _______________________________________________ >> mpi-forum mailing list >> mpi-forum(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >> > > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji > _______________________________________________ > 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] MPI "Allocate receive" proposal
by Pavan Balaji 26 Aug '13

26 Aug '13

Jeff, This should probably be moved to the point-to-point WG, but I'll comment on this thread for now. Such a proposal was brought up at the very early stage of the MPI-2.1 process (I don't remember by whom). The main concern at that point was what happens when the memory allocation fails. Will the receive fail? What happens to the sent message? Can you work these details into your slides? -- Pavan On 08/26/2013 10:10 AM, Jeff Squyres (jsquyres) wrote: > Dave Goodell and I have a proposal that we've socialized a bit around with other Forum members, and would now like larger Forum feedback. I'll be presenting the attached slides on the concept of an "allocate receive" in Madrid (3:30-4pm on Thursday). > > There's no text or ticket yet; this is an idea that we want to get feedback on before working up a full proposal. > > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

MPI "Allocate receive" proposal
by Jeff Squyres (jsquyres) 26 Aug '13

26 Aug '13

Dave Goodell and I have a proposal that we've socialized a bit around with other Forum members, and would now like larger Forum feedback. I'll be presenting the attached slides on the concept of an "allocate receive" in Madrid (3:30-4pm on Thursday). There's no text or ticket yet; this is an idea that we want to get feedback on before working up a full proposal. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Agenda for MPI-Forum September Meeting in Madrid
by Schulz, Martin 23 Aug '13

23 Aug '13

Hi all, I updated the agenda for the September meeting based on the input I have received. Please take a look at it, especially if you asked for time, and let me know if I missed anything. As of now, I only got agenda items for two days and almost no WG request (which, I guess, is not too surprising). I'll therefore leave Wednesday as an open discussion day that WGs can use to self-organize meetings. The meeting space will be available, although we won't cater lunch. I have listed all tickets that will be (officially) read either as part of MPI next or MPI 3 errata and I consider those announced as required. The WG chairs should make sure, though, that the latest versions of the respective proposals texts are available from the tickets by the deadline next Tuesday night PDT (9am Wednesday Madrid time). If there are any other official readings or errata items, please send them to me as soon as you can, but latest by that deadline. As for the quorum: based on the initial feedback that I have gotten, we will have quorum at this meeting, so votes will count, but this is of course tentative and depends on the actual attendance at the meeting. Please let me know if there are any questions or comments, See you in Madrid, Martin ________________________________________________________________________ Martin Schulz, schulzm(a)llnl.gov, http://people.llnl.gov/schulzm CASC @ Lawrence Livermore National Laboratory, Livermore, USA

1 0