November 2017 - mpiwg-persistence - mailman3.cels.anl.gov

Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29)
by Anthony Skjellum 30 Nov '17

30 Nov '17

the wording seems OK to me Anthony Skjellum, PhD Affiliate Faculty, Computer Science and Software Engineering at Auburn University 205-807-4968 (cell) Preferred e-mails for me are skjellum(a)gmail.com or tony-skjellum(a)utc.edu ________________________________ From: mpiwg-persistence <mpiwg-persistence-bounces(a)lists.mpi-forum.org> on behalf of HOLMES Daniel <d.holmes(a)epcc.ed.ac.uk> Sent: Thursday, November 30, 2017 5:18 AM To: MPI WG Persistence working Group Cc: Martin Schulz; Anthony Skjellum Subject: Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29) Hi all, I’ve just pushed this text change as is to the Persistence WG github - so it is now included on our pull request. Do we wish to word-smith the exact phraseology at all? I’m happy with it as is, but I can easily make a wording change if we decide that’s needed or wanted. Once we have WG sign-off on the wording, I’ll rebuild the PDF. Cheers, Dan. On 30 Nov 2017, at 01:43, Anthony Skjellum <skjellum(a)auburn.edu<mailto:[email protected]>> wrote: Martin, OK, thank you. That makes perfect sense.-Tony Anthony Skjellum, PhD Affiliate Faculty, Computer Science and Software Engineering at Auburn University 205-807-4968 (cell) Preferred e-mails for me are skjellum(a)gmail.com<mailto:[email protected]> or tony-skjellum(a)utc.edu<mailto:[email protected]> ________________________________ From: mpiwg-persistence <mpiwg-persistence-bounces(a)lists.mpi-forum.org<mailto:[email protected]>> on behalf of Martin Schulz <gu58niw(a)mytum.de<mailto:[email protected]>> Sent: Wednesday, November 29, 2017 5:04 PM To: Anthony Skjellum Cc: MPI WG Persistence working Group Subject: Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29) Hi Tony, On 29. Nov 2017, at 23:58, Anthony Skjellum <skjellum(a)gmail.com<mailto:[email protected]>> wrote: Can this be a "ticket 0" change? No, I think this is clearly more than a ticket 0 change, but as Dan suggested, it should pass the no/no vote test. Martin On Wed, Nov 29, 2017 at 2:56 PM, Martin Schulz <gu58niw(a)mytum.de<mailto:[email protected]>>wrote: Hi Dan, all, Just my personal view: the that’s how I always interpreted the statement anyway, but I can see how this could be misinterpreted. Hence, adding the clarification is IMHO a good move. Just my 2c, Martin On 29. Nov 2017, at 11:07, HOLMES Daniel <d.holmes(a)epcc.ed.ac.uk<mailto:[email protected]>> wrote: Hi Persistence WG members, We have a suggested change to the wording of the MPI Standard text that we will propose to the MPI Forum next week regarding persistent collective operations. Once initialized, persistent collective operations can be started in any order independently of initialization order +and the order can differ among ranks in the communicator Most of this sentence already appears in our proposed text - the suggestion is to add the line beginning with “+”. I think this is a good suggestion - we should make this point absolutely clear. We should word-smith the exact phrase (e.g. during the meeting later today?) and then include it in our formal reading next week by holding a “no-no” vote. Cheers, Dan. Begin forwarded message: From: KAWASHIMA Takahiro <notifications(a)github.com<mailto:[email protected]>> Subject: Re: [mpi-forum/mpi-standard] Persistent collectives 2017 (#29) Date: 29 November 2017 at 06:09:20 GMT To: mpi-forum/mpi-standard <mpi-standard(a)noreply.github.com<mailto:[email protected]>> Cc: Dan Holmes <dholmes(a)epcc.ed.ac.uk<mailto:[email protected]>>, Mention <mention(a)noreply.github.com<mailto:[email protected]>> Reply-To: mpi-forum/mpi-standard <reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755(a)reply.github.com<mailto:reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755@reply.github.com>> @dholmes-epcc-ed-ac-uk<https://github.com/dholmes-epcc-ed-ac-uk>, With the existing text, I think many people can understand that it is saying (at least) "requests can be started using MPI_START or MPI_STARTALL in any order independently of initialization order", because of analogy to the point-to-point communication. So I think your suggested text is unnecessary. But some people may not understand that it is also saying "the order of starting requests can differ among ranks in the communicator", because (1) the matching rule for persistent collective operations is different from the matching rule for point-to-point communication so that analogy to the point-to-point communication is not applied and (2) the order of nonblocking collective calls must be same among ranks in the communicator. How about the sentence below? I'm not good at English. Please correct it if it is not good. Once initialized, persistent collective operations can be started in any order independently of initialization order and the order can differ among ranks in the communicator If anyone else thinks the existing text is sufficient, I won't request the change. Probably my English reading skill is low. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<https://github.com/mpi-forum/mpi-standard/pull/29#issuecomment-347762876>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AMf8SiG0KfwGmYxNhaFoTBv_J…>. The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. _______________________________________________ mpiwg-persistence mailing list mpiwg-persistence(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence _______________________________________________ mpiwg-persistence mailing list mpiwg-persistence(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence -- Anthony Skjellum, PhD skjellum(a)gmail.com<mailto:[email protected]> Cell: +1-205-807-4968 _______________________________________________ mpiwg-persistence mailing list mpiwg-persistence(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence

1 0

Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29)
by HOLMES Daniel 30 Nov '17

30 Nov '17

Hi all, I’ve just pushed this text change as is to the Persistence WG github - so it is now included on our pull request. Do we wish to word-smith the exact phraseology at all? I’m happy with it as is, but I can easily make a wording change if we decide that’s needed or wanted. Once we have WG sign-off on the wording, I’ll rebuild the PDF. Cheers, Dan. On 30 Nov 2017, at 01:43, Anthony Skjellum <skjellum(a)auburn.edu<mailto:[email protected]>> wrote: Martin, OK, thank you. That makes perfect sense.-Tony Anthony Skjellum, PhD Affiliate Faculty, Computer Science and Software Engineering at Auburn University 205-807-4968 (cell) Preferred e-mails for me are skjellum(a)gmail.com<mailto:[email protected]> or tony-skjellum(a)utc.edu<mailto:[email protected]> ________________________________ From: mpiwg-persistence <mpiwg-persistence-bounces(a)lists.mpi-forum.org<mailto:[email protected]>> on behalf of Martin Schulz <gu58niw(a)mytum.de<mailto:[email protected]>> Sent: Wednesday, November 29, 2017 5:04 PM To: Anthony Skjellum Cc: MPI WG Persistence working Group Subject: Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29) Hi Tony, On 29. Nov 2017, at 23:58, Anthony Skjellum <skjellum(a)gmail.com<mailto:[email protected]>> wrote: Can this be a "ticket 0" change? No, I think this is clearly more than a ticket 0 change, but as Dan suggested, it should pass the no/no vote test. Martin On Wed, Nov 29, 2017 at 2:56 PM, Martin Schulz <gu58niw(a)mytum.de<mailto:[email protected]>>wrote: Hi Dan, all, Just my personal view: the that’s how I always interpreted the statement anyway, but I can see how this could be misinterpreted. Hence, adding the clarification is IMHO a good move. Just my 2c, Martin On 29. Nov 2017, at 11:07, HOLMES Daniel <d.holmes(a)epcc.ed.ac.uk<mailto:[email protected]>> wrote: Hi Persistence WG members, We have a suggested change to the wording of the MPI Standard text that we will propose to the MPI Forum next week regarding persistent collective operations. Once initialized, persistent collective operations can be started in any order independently of initialization order +and the order can differ among ranks in the communicator Most of this sentence already appears in our proposed text - the suggestion is to add the line beginning with “+”. I think this is a good suggestion - we should make this point absolutely clear. We should word-smith the exact phrase (e.g. during the meeting later today?) and then include it in our formal reading next week by holding a “no-no” vote. Cheers, Dan. Begin forwarded message: From: KAWASHIMA Takahiro <notifications(a)github.com<mailto:[email protected]>> Subject: Re: [mpi-forum/mpi-standard] Persistent collectives 2017 (#29) Date: 29 November 2017 at 06:09:20 GMT To: mpi-forum/mpi-standard <mpi-standard(a)noreply.github.com<mailto:[email protected]>> Cc: Dan Holmes <dholmes(a)epcc.ed.ac.uk<mailto:[email protected]>>, Mention <mention(a)noreply.github.com<mailto:[email protected]>> Reply-To: mpi-forum/mpi-standard <reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755(a)reply.github.com<mailto:reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755@reply.github.com>> @dholmes-epcc-ed-ac-uk<https://github.com/dholmes-epcc-ed-ac-uk>, With the existing text, I think many people can understand that it is saying (at least) "requests can be started using MPI_START or MPI_STARTALL in any order independently of initialization order", because of analogy to the point-to-point communication. So I think your suggested text is unnecessary. But some people may not understand that it is also saying "the order of starting requests can differ among ranks in the communicator", because (1) the matching rule for persistent collective operations is different from the matching rule for point-to-point communication so that analogy to the point-to-point communication is not applied and (2) the order of nonblocking collective calls must be same among ranks in the communicator. How about the sentence below? I'm not good at English. Please correct it if it is not good. Once initialized, persistent collective operations can be started in any order independently of initialization order and the order can differ among ranks in the communicator If anyone else thinks the existing text is sufficient, I won't request the change. Probably my English reading skill is low. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<https://github.com/mpi-forum/mpi-standard/pull/29#issuecomment-347762876>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AMf8SiG0KfwGmYxNhaFoTBv_J…>. The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. _______________________________________________ mpiwg-persistence mailing list mpiwg-persistence(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence _______________________________________________ mpiwg-persistence mailing list mpiwg-persistence(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence -- Anthony Skjellum, PhD skjellum(a)gmail.com<mailto:[email protected]> Cell: +1-205-807-4968 _______________________________________________ mpiwg-persistence mailing list mpiwg-persistence(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.

1 0

Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29)
by Anthony Skjellum 29 Nov '17

29 Nov '17

Martin, OK, thank you. That makes perfect sense.-Tony Anthony Skjellum, PhD Affiliate Faculty, Computer Science and Software Engineering at Auburn University 205-807-4968 (cell) Preferred e-mails for me are skjellum(a)gmail.com or tony-skjellum(a)utc.edu ________________________________ From: mpiwg-persistence <mpiwg-persistence-bounces(a)lists.mpi-forum.org> on behalf of Martin Schulz <gu58niw(a)mytum.de> Sent: Wednesday, November 29, 2017 5:04 PM To: Anthony Skjellum Cc: MPI WG Persistence working Group Subject: Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29) Hi Tony, On 29. Nov 2017, at 23:58, Anthony Skjellum <skjellum(a)gmail.com<mailto:[email protected]>> wrote: Can this be a "ticket 0" change? No, I think this is clearly more than a ticket 0 change, but as Dan suggested, it should pass the no/no vote test. Martin On Wed, Nov 29, 2017 at 2:56 PM, Martin Schulz <gu58niw(a)mytum.de<mailto:[email protected]>> wrote: Hi Dan, all, Just my personal view: the that’s how I always interpreted the statement anyway, but I can see how this could be misinterpreted. Hence, adding the clarification is IMHO a good move. Just my 2c, Martin On 29. Nov 2017, at 11:07, HOLMES Daniel <d.holmes(a)epcc.ed.ac.uk<mailto:[email protected]>> wrote: Hi Persistence WG members, We have a suggested change to the wording of the MPI Standard text that we will propose to the MPI Forum next week regarding persistent collective operations. Once initialized, persistent collective operations can be started in any order independently of initialization order +and the order can differ among ranks in the communicator Most of this sentence already appears in our proposed text - the suggestion is to add the line beginning with “+”. I think this is a good suggestion - we should make this point absolutely clear. We should word-smith the exact phrase (e.g. during the meeting later today?) and then include it in our formal reading next week by holding a “no-no” vote. Cheers, Dan. Begin forwarded message: From: KAWASHIMA Takahiro <notifications(a)github.com<mailto:[email protected]>> Subject: Re: [mpi-forum/mpi-standard] Persistent collectives 2017 (#29) Date: 29 November 2017 at 06:09:20 GMT To: mpi-forum/mpi-standard <mpi-standard(a)noreply.github.com<mailto:[email protected]>> Cc: Dan Holmes <dholmes(a)epcc.ed.ac.uk<mailto:[email protected]>>, Mention <mention(a)noreply.github.com<mailto:[email protected]>> Reply-To: mpi-forum/mpi-standard <reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755(a)reply.github.com<mailto:reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755@reply.github.com>> @dholmes-epcc-ed-ac-uk<https://github.com/dholmes-epcc-ed-ac-uk>, With the existing text, I think many people can understand that it is saying (at least) "requests can be started using MPI_START or MPI_STARTALL in any order independently of initialization order", because of analogy to the point-to-point communication. So I think your suggested text is unnecessary. But some people may not understand that it is also saying "the order of starting requests can differ among ranks in the communicator", because (1) the matching rule for persistent collective operations is different from the matching rule for point-to-point communication so that analogy to the point-to-point communication is not applied and (2) the order of nonblocking collective calls must be same among ranks in the communicator. How about the sentence below? I'm not good at English. Please correct it if it is not good. Once initialized, persistent collective operations can be started in any order independently of initialization order and the order can differ among ranks in the communicator If anyone else thinks the existing text is sufficient, I won't request the change. Probably my English reading skill is low. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<https://github.com/mpi-forum/mpi-standard/pull/29#issuecomment-347762876>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AMf8SiG0KfwGmYxNhaFoTBv_J…>. The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. _______________________________________________ mpiwg-persistence mailing list mpiwg-persistence(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence _______________________________________________ mpiwg-persistence mailing list mpiwg-persistence(a)lists.mpi-forum.org<mailto:[email protected]> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence -- Anthony Skjellum, PhD skjellum(a)gmail.com<mailto:[email protected]> Cell: +1-205-807-4968

1 0

Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29)
by Martin Schulz 29 Nov '17

29 Nov '17

Hi Tony, > On 29. Nov 2017, at 23:58, Anthony Skjellum <skjellum(a)gmail.com> wrote: > > Can this be a "ticket 0" change? No, I think this is clearly more than a ticket 0 change, but as Dan suggested, it should pass the no/no vote test. Martin > > On Wed, Nov 29, 2017 at 2:56 PM, Martin Schulz <gu58niw(a)mytum.de <mailto:[email protected]>> wrote: > Hi Dan, all, > > Just my personal view: the that’s how I always interpreted the statement anyway, but I can see how this could be misinterpreted. Hence, adding the clarification is IMHO a good move. > > Just my 2c, > > Martin > > >> On 29. Nov 2017, at 11:07, HOLMES Daniel <d.holmes(a)epcc.ed.ac.uk <mailto:[email protected]>> wrote: >> >> Hi Persistence WG members, >> >> We have a suggested change to the wording of the MPI Standard text that we will propose to the MPI Forum next week regarding persistent collective operations. >> >>> Once initialized, >>> persistent collective operations can be started in any order >>> independently of initialization order >>> +and the order can differ among ranks in the communicator >> >> Most of this sentence already appears in our proposed text - the suggestion is to add the line beginning with “+”. >> >> I think this is a good suggestion - we should make this point absolutely clear. We should word-smith the exact phrase (e.g. during the meeting later today?) and then include it in our formal reading next week by holding a “no-no” vote. >> >> Cheers, >> Dan. >> >>> Begin forwarded message: >>> >>> From: KAWASHIMA Takahiro <notifications(a)github.com <mailto:[email protected]>> >>> Subject: Re: [mpi-forum/mpi-standard] Persistent collectives 2017 (#29) >>> Date: 29 November 2017 at 06:09:20 GMT >>> To: mpi-forum/mpi-standard <mpi-standard(a)noreply.github.com <mailto:[email protected]>> >>> Cc: Dan Holmes <dholmes(a)epcc.ed.ac.uk <mailto:[email protected]>>, Mention <mention(a)noreply.github.com <mailto:[email protected]>> >>> Reply-To: mpi-forum/mpi-standard <reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755(a)reply.github.com <mailto:reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755@reply.github.com>> >>> >>> @dholmes-epcc-ed-ac-uk <https://github.com/dholmes-epcc-ed-ac-uk>, >>> With the existing text, I think many people can understand that it is saying (at least) "requests can be started using MPI_START or MPI_STARTALL in any order independently of initialization order", because of analogy to the point-to-point communication. So I think your suggested text is unnecessary. >>> But some people may not understand that it is also saying "the order of starting requests can differ among ranks in the communicator", because (1) the matching rule for persistent collective operations is different from the matching rule for point-to-point communication so that analogy to the point-to-point communication is not applied and (2) the order of nonblocking collective calls must be same among ranks in the communicator. >>> >>> How about the sentence below? I'm not good at English. Please correct it if it is not good. >>> >>> Once initialized, >>> persistent collective operations can be started in any order >>> independently of initialization order >>> and the order can differ among ranks in the communicator >>> >>> If anyone else thinks the existing text is sufficient, I won't request the change. Probably my English reading skill is low. >>> >>> — >>> You are receiving this because you were mentioned. >>> Reply to this email directly, view it on GitHub <https://github.com/mpi-forum/mpi-standard/pull/29#issuecomment-347762876>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AMf8SiG0KfwGmYxNhaFoTBv_J…>. >>> >> >> The University of Edinburgh is a charitable body, registered in >> Scotland, with registration number SC005336. >> _______________________________________________ >> mpiwg-persistence mailing list >> mpiwg-persistence(a)lists.mpi-forum.org <mailto:[email protected]> >> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence <https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence> > > > _______________________________________________ > mpiwg-persistence mailing list > mpiwg-persistence(a)lists.mpi-forum.org <mailto:[email protected]> > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence <https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence> > > > > > -- > Anthony Skjellum, PhD > skjellum(a)gmail.com <mailto:[email protected]> > Cell: +1-205-807-4968 > >

1 0

Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29)
by Anthony Skjellum 29 Nov '17

29 Nov '17

Can this be a "ticket 0" change? On Wed, Nov 29, 2017 at 2:56 PM, Martin Schulz <gu58niw(a)mytum.de> wrote: > Hi Dan, all, > > Just my personal view: the that’s how I always interpreted the statement > anyway, but I can see how this could be misinterpreted. Hence, adding the > clarification is IMHO a good move. > > Just my 2c, > > Martin > > > On 29. Nov 2017, at 11:07, HOLMES Daniel <d.holmes(a)epcc.ed.ac.uk> wrote: > > Hi Persistence WG members, > > We have a suggested change to the wording of the MPI Standard text that we > will propose to the MPI Forum next week regarding persistent collective > operations. > > Once initialized, > persistent collective operations can be started in any order > independently of initialization order > +and the order can differ among ranks in the communicator > > > Most of this sentence already appears in our proposed text - the > suggestion is to add the line beginning with “+”. > > I think this is a good suggestion - we should make this point absolutely > clear. We should word-smith the exact phrase (e.g. during the meeting later > today?) and then include it in our formal reading next week by holding a > “no-no” vote. > > Cheers, > Dan. > > Begin forwarded message: > > *From: *KAWASHIMA Takahiro <notifications(a)github.com> > *Subject: **Re: [mpi-forum/mpi-standard] Persistent collectives 2017 > (#29)* > *Date: *29 November 2017 at 06:09:20 GMT > *To: *mpi-forum/mpi-standard <mpi-standard(a)noreply.github.com> > *Cc: *Dan Holmes <dholmes(a)epcc.ed.ac.uk>, Mention < > mention(a)noreply.github.com> > *Reply-To: *mpi-forum/mpi-standard <reply+00c7fc4ac57b95d8d1ec9f08428087 > 76a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755(a)reply.github.com> > > @dholmes-epcc-ed-ac-uk <https://github.com/dholmes-epcc-ed-ac-uk>, > With the existing text, I think many people can understand that it is > saying (at least) "*requests can be started using MPI_START or > MPI_STARTALL in any order independently of initialization order*", > because of analogy to the point-to-point communication. So I think your > suggested text is unnecessary. > But some people may *not* understand that it is also saying "*the order > of starting requests can differ among ranks in the communicator*", > because (1) the matching rule for persistent collective operations is > different from the matching rule for point-to-point communication so that > analogy to the point-to-point communication is not applied and (2) the > order of nonblocking collective calls must be same among ranks in the > communicator. > > How about the sentence below? I'm not good at English. Please correct it > if it is not good. > > Once initialized, > persistent collective operations can be started in any order > independently of initialization order > and the order can differ among ranks in the communicator > > If anyone else thinks the existing text is sufficient, I won't request the > change. Probably my English reading skill is low. > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <https://github.com/mpi-forum/mpi-standard/pull/29#issuecomment-347762876>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AMf8SiG0KfwGmYxNhaFoTBv_J…> > . > > > The University of Edinburgh is a charitable body, registered in > Scotland, with registration number SC005336. > _______________________________________________ > mpiwg-persistence mailing list > mpiwg-persistence(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence > > > > _______________________________________________ > mpiwg-persistence mailing list > mpiwg-persistence(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence > > -- Anthony Skjellum, PhD skjellum(a)gmail.com Cell: +1-205-807-4968

1 0

Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29)
by Anthony Skjellum 29 Nov '17

29 Nov '17

This is a good suggestion, agreed. On Wed, Nov 29, 2017 at 5:07 AM, HOLMES Daniel <d.holmes(a)epcc.ed.ac.uk> wrote: > Hi Persistence WG members, > > We have a suggested change to the wording of the MPI Standard text that we > will propose to the MPI Forum next week regarding persistent collective > operations. > > Once initialized, > persistent collective operations can be started in any order > independently of initialization order > +and the order can differ among ranks in the communicator > > > Most of this sentence already appears in our proposed text - the > suggestion is to add the line beginning with “+”. > > I think this is a good suggestion - we should make this point absolutely > clear. We should word-smith the exact phrase (e.g. during the meeting later > today?) and then include it in our formal reading next week by holding a > “no-no” vote. > > Cheers, > Dan. > > Begin forwarded message: > > *From: *KAWASHIMA Takahiro <notifications(a)github.com> > *Subject: **Re: [mpi-forum/mpi-standard] Persistent collectives 2017 > (#29)* > *Date: *29 November 2017 at 06:09:20 GMT > *To: *mpi-forum/mpi-standard <mpi-standard(a)noreply.github.com> > *Cc: *Dan Holmes <dholmes(a)epcc.ed.ac.uk>, Mention < > mention(a)noreply.github.com> > *Reply-To: *mpi-forum/mpi-standard <reply+00c7fc4ac57b95d8d1ec9f08428087 > 76a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755(a)reply.github.com> > > @dholmes-epcc-ed-ac-uk <https://github.com/dholmes-epcc-ed-ac-uk>, > With the existing text, I think many people can understand that it is > saying (at least) "*requests can be started using MPI_START or > MPI_STARTALL in any order independently of initialization order*", > because of analogy to the point-to-point communication. So I think your > suggested text is unnecessary. > But some people may *not* understand that it is also saying "*the order > of starting requests can differ among ranks in the communicator*", > because (1) the matching rule for persistent collective operations is > different from the matching rule for point-to-point communication so that > analogy to the point-to-point communication is not applied and (2) the > order of nonblocking collective calls must be same among ranks in the > communicator. > > How about the sentence below? I'm not good at English. Please correct it > if it is not good. > > Once initialized, > persistent collective operations can be started in any order > independently of initialization order > and the order can differ among ranks in the communicator > > If anyone else thinks the existing text is sufficient, I won't request the > change. Probably my English reading skill is low. > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <https://github.com/mpi-forum/mpi-standard/pull/29#issuecomment-347762876>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AMf8SiG0KfwGmYxNhaFoTBv_J…> > . > > > > The University of Edinburgh is a charitable body, registered in > Scotland, with registration number SC005336. > > -- Anthony Skjellum, PhD skjellum(a)gmail.com Cell: +1-205-807-4968

1 0

Re: [mpiwg-persistence] [mpi-forum/mpi-standard] Persistent collectives 2017 (#29)
by Martin Schulz 29 Nov '17

29 Nov '17

Hi Dan, all, Just my personal view: the that’s how I always interpreted the statement anyway, but I can see how this could be misinterpreted. Hence, adding the clarification is IMHO a good move. Just my 2c, Martin > On 29. Nov 2017, at 11:07, HOLMES Daniel <d.holmes(a)epcc.ed.ac.uk> wrote: > > Hi Persistence WG members, > > We have a suggested change to the wording of the MPI Standard text that we will propose to the MPI Forum next week regarding persistent collective operations. > >> Once initialized, >> persistent collective operations can be started in any order >> independently of initialization order >> +and the order can differ among ranks in the communicator > > Most of this sentence already appears in our proposed text - the suggestion is to add the line beginning with “+”. > > I think this is a good suggestion - we should make this point absolutely clear. We should word-smith the exact phrase (e.g. during the meeting later today?) and then include it in our formal reading next week by holding a “no-no” vote. > > Cheers, > Dan. > >> Begin forwarded message: >> >> From: KAWASHIMA Takahiro <notifications(a)github.com <mailto:[email protected]>> >> Subject: Re: [mpi-forum/mpi-standard] Persistent collectives 2017 (#29) >> Date: 29 November 2017 at 06:09:20 GMT >> To: mpi-forum/mpi-standard <mpi-standard(a)noreply.github.com <mailto:[email protected]>> >> Cc: Dan Holmes <dholmes(a)epcc.ed.ac.uk <mailto:[email protected]>>, Mention <mention(a)noreply.github.com <mailto:[email protected]>> >> Reply-To: mpi-forum/mpi-standard <reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755(a)reply.github.com <mailto:reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755@reply.github.com>> >> >> @dholmes-epcc-ed-ac-uk <https://github.com/dholmes-epcc-ed-ac-uk>, >> With the existing text, I think many people can understand that it is saying (at least) "requests can be started using MPI_START or MPI_STARTALL in any order independently of initialization order", because of analogy to the point-to-point communication. So I think your suggested text is unnecessary. >> But some people may not understand that it is also saying "the order of starting requests can differ among ranks in the communicator", because (1) the matching rule for persistent collective operations is different from the matching rule for point-to-point communication so that analogy to the point-to-point communication is not applied and (2) the order of nonblocking collective calls must be same among ranks in the communicator. >> >> How about the sentence below? I'm not good at English. Please correct it if it is not good. >> >> Once initialized, >> persistent collective operations can be started in any order >> independently of initialization order >> and the order can differ among ranks in the communicator >> >> If anyone else thinks the existing text is sufficient, I won't request the change. Probably my English reading skill is low. >> >> — >> You are receiving this because you were mentioned. >> Reply to this email directly, view it on GitHub <https://github.com/mpi-forum/mpi-standard/pull/29#issuecomment-347762876>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AMf8SiG0KfwGmYxNhaFoTBv_J…>. >> > > The University of Edinburgh is a charitable body, registered in > Scotland, with registration number SC005336. > _______________________________________________ > mpiwg-persistence mailing list > mpiwg-persistence(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence

1 0

Fwd: [mpi-forum/mpi-standard] Persistent collectives 2017 (#29)
by HOLMES Daniel 29 Nov '17

29 Nov '17

Hi Persistence WG members, We have a suggested change to the wording of the MPI Standard text that we will propose to the MPI Forum next week regarding persistent collective operations. Once initialized, persistent collective operations can be started in any order independently of initialization order +and the order can differ among ranks in the communicator Most of this sentence already appears in our proposed text - the suggestion is to add the line beginning with “+”. I think this is a good suggestion - we should make this point absolutely clear. We should word-smith the exact phrase (e.g. during the meeting later today?) and then include it in our formal reading next week by holding a “no-no” vote. Cheers, Dan. Begin forwarded message: From: KAWASHIMA Takahiro <notifications(a)github.com<mailto:[email protected]>> Subject: Re: [mpi-forum/mpi-standard] Persistent collectives 2017 (#29) Date: 29 November 2017 at 06:09:20 GMT To: mpi-forum/mpi-standard <mpi-standard(a)noreply.github.com<mailto:[email protected]>> Cc: Dan Holmes <dholmes(a)epcc.ed.ac.uk<mailto:[email protected]>>, Mention <mention(a)noreply.github.com<mailto:[email protected]>> Reply-To: mpi-forum/mpi-standard <reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755(a)reply.github.com<mailto:reply+00c7fc4ac57b95d8d1ec9f0842808776a6bf83a41aa2b15e92cf000000011636110f92a169ce0dd58755@reply.github.com>> @dholmes-epcc-ed-ac-uk<https://github.com/dholmes-epcc-ed-ac-uk>, With the existing text, I think many people can understand that it is saying (at least) "requests can be started using MPI_START or MPI_STARTALL in any order independently of initialization order", because of analogy to the point-to-point communication. So I think your suggested text is unnecessary. But some people may not understand that it is also saying "the order of starting requests can differ among ranks in the communicator", because (1) the matching rule for persistent collective operations is different from the matching rule for point-to-point communication so that analogy to the point-to-point communication is not applied and (2) the order of nonblocking collective calls must be same among ranks in the communicator. How about the sentence below? I'm not good at English. Please correct it if it is not good. Once initialized, persistent collective operations can be started in any order independently of initialization order and the order can differ among ranks in the communicator If anyone else thinks the existing text is sufficient, I won't request the change. Probably my English reading skill is low. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<https://github.com/mpi-forum/mpi-standard/pull/29#issuecomment-347762876>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AMf8SiG0KfwGmYxNhaFoTBv_J…>. The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.

1 0

Re: [mpiwg-persistence] Freeing collective requests
by Anthony Skjellum 17 Nov '17

17 Nov '17

Dan, thanks for doing all this hard work. I agree will read the whole proposal, and the document is a diff against MPI 3.1, so we will not be hiding anything from the baseline :-) I will review this weekend, but I am going to announce the reading now. Thanks, Tony On Fri, Nov 17, 2017 at 4:00 PM, HOLMES Daniel <d.holmes(a)epcc.ed.ac.uk> wrote: > Hi all, > > I have added a clarification to the point-to-point chapter (where > MPI_REQUEST_FREE is defined) to state that collective requests must not be > freed while they are active. > > I have pushed these changes to Github and produced a new PDF of the MPI > Standard (attached to comments on both the issue and the pull request). > I have added a comment in Github (on both the issue and the pull request) > summarising the current thinking regarding “collective free” semantics and > “block until really deallocated” semantics. > > Useful links: > https://github.com/mpi-forum/mpi-issues/issues/25 > https://github.com/mpi-forum/mpi-standard/pull/29 > https://github.com/mpi-forum/mpi-standard/files/1483740/mpi32-report.pdf > > I think this means we are ready for the next face-to-face MPI Forum > meeting. > > We made quite a few small changes since the version announced for the last > meeting. We should re-read the whole proposal, rather than try to get away > with a “no no vote” followed by a 1st vote. > > If there are other tasks to complete before we announce our reading > (deadline Monday 20th Nov), please let me know by replying to this email > list ASAP. > > Cheers, > Dan. > > The University of Edinburgh is a charitable body, registered in > Scotland, with registration number SC005336. > > _______________________________________________ > mpiwg-persistence mailing list > mpiwg-persistence(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-persistence > > -- Anthony Skjellum, PhD skjellum(a)gmail.com Cell: +1-205-807-4968

1 0

Freeing collective requests
by HOLMES Daniel 17 Nov '17

17 Nov '17

Hi all, I have added a clarification to the point-to-point chapter (where MPI_REQUEST_FREE is defined) to state that collective requests must not be freed while they are active. I have pushed these changes to Github and produced a new PDF of the MPI Standard (attached to comments on both the issue and the pull request). I have added a comment in Github (on both the issue and the pull request) summarising the current thinking regarding “collective free” semantics and “block until really deallocated” semantics. Useful links: https://github.com/mpi-forum/mpi-issues/issues/25 https://github.com/mpi-forum/mpi-standard/pull/29 https://github.com/mpi-forum/mpi-standard/files/1483740/mpi32-report.pdf I think this means we are ready for the next face-to-face MPI Forum meeting. We made quite a few small changes since the version announced for the last meeting. We should re-read the whole proposal, rather than try to get away with a “no no vote” followed by a 1st vote. If there are other tasks to complete before we announce our reading (deadline Monday 20th Nov), please let me know by replying to this email list ASAP. Cheers, Dan. The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.

1 0