December 2014 - mpiwg-coll - mailman3.cels.anl.gov

Re: [mpiwg-coll] deprecating MPI_GRAPH_CREATE?
by Torsten Hoefler 02 Dec '14

02 Dec '14

Is there an alternative? :-) Torsten PS: I will stop spamming the list with this topic now. On 12/02/2014 08:41 PM, Jeff Hammond wrote: > Does the non-scalable nature of v-collectives prevent people from using them? > > :-) > > Jeff > > On Tue, Dec 2, 2014 at 11:38 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: >> I consider this risk as low because it says it in the standard that it's >> non-scalable as well as in any recent documentation. Yes, some people will >> write bad codes. >> >> All the Best, >> Torsten >> >> >> On 12/02/2014 08:28 PM, Jeff Hammond wrote: >>> >>> I created ticket 462 as a placeholder for whenever someone has time to >>> work on this. >>> >>> MPI_GRAPH_CREATE does harm to users because they might use it. >>> Somebody might naively think that distributed graphs are more >>> complicated to use than replicated ones... >>> >>> Jeff >>> >>> On Tue, Dec 2, 2014 at 1:57 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: >>>> >>>> I am unfortunately too busy these days to drive it (a clean removal will >>>> touch several places). It is ugly but it doesn't do any direct harm to >>>> the >>>> standard or its users. >>>> >>>> Best, >>>> Torsten >>>> >>>> >>>> On 01.12.2014 17:02, Jeff Hammond wrote: >>>>> >>>>> >>>>> But you're not going to write that proposal? You seem more qualified >>>>> than >>>>> most... >>>>> >>>>> Jeff >>>>> >>>>> Sent from my iPhone >>>>> >>>>>> On Dec 1, 2014, at 7:51 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: >>>>>> >>>>>> Jeff, >>>>>> >>>>>> There is no such proposal yet (afaik), I'd support a carefully crafted >>>>>> one and I think I wouldn't be alone. >>>>>> >>>>>> Thanks & Best, >>>>>> Torsten >>>>>> >>>>>>> On 01.12.2014 06:41, Jeff Hammond wrote: >>>>>>> Regarding http://lists.mpi-forum.org/mpiwg-coll/2014/03/0289.php, is >>>>>>> there a proposal to deprecate MPI_GRAPH_CREATE? It seems that there >>>>>>> are many issues with it and they are deliberately not getting fixed >>>>>>> because a subset of the issues with it make it too horrible to fix. >>>>>>> >>>>>>> I find this situation distasteful as it leads to a poor user >>>>>>> experience. Let's do the right thing and put MPI_GRAPH_CREATE out of >>>>>>> our misery sooner rather than later. >>>>>>> >>>>>>> Thanks, >>>>>>> >>>>>>> Jeff >>>>>> >>>>>> >>>>>> >>>>>> -- >>>>>> ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- >>>>>> Torsten Hoefler | Assistant Professor >>>>>> Dept. of Computer Science | ETH Zürich >>>>>> Universitätsstrasse 6 | Zurich-8092, Switzerland >>>>>> CAB E 64.1 | Phone: +41 76 309 79 29 >>>>>> _______________________________________________ >>>>>> mpiwg-coll mailing list >>>>>> mpiwg-coll(a)lists.mpi-forum.org >>>>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll >>>>> >>>>> >>>>> _______________________________________________ >>>>> mpiwg-coll mailing list >>>>> mpiwg-coll(a)lists.mpi-forum.org >>>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll >>>>> >>>> >>>> -- >>>> ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- >>>> Torsten Hoefler | Assistant Professor >>>> Dept. of Computer Science | ETH Zürich >>>> Universitätsstrasse 6 | Zurich-8092, Switzerland >>>> CAB E 64.1 | Phone: +41 76 309 79 29 >>>> _______________________________________________ >>>> mpiwg-coll mailing list >>>> mpiwg-coll(a)lists.mpi-forum.org >>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll >>> >>> >>> >>> >> >> -- >> ### qreharg rug ebs fv crryF ------------- http://www.unixer.de/ ----- >> >> Torsten Hoefler | Assistant Professor >> Dept. of Computer Science | ETH Zürich >> Universitätsstrasse 6 | Zurich-8092, Switzerland >> CAB E 64.1 | Phone: +41 76 309 79 29 >> _______________________________________________ >> mpiwg-coll mailing list >> mpiwg-coll(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll > > > -- ### qreharg rug ebs fv crryF ------------- http://www.unixer.de/ ----- Torsten Hoefler | Assistant Professor Dept. of Computer Science | ETH Zürich Universitätsstrasse 6 | Zurich-8092, Switzerland CAB E 64.1 | Phone: +41 76 309 79 29

1 0

Re: [mpiwg-coll] deprecating MPI_GRAPH_CREATE?
by Jeff Hammond 02 Dec '14

02 Dec '14

Does the non-scalable nature of v-collectives prevent people from using them? :-) Jeff On Tue, Dec 2, 2014 at 11:38 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: > I consider this risk as low because it says it in the standard that it's > non-scalable as well as in any recent documentation. Yes, some people will > write bad codes. > > All the Best, > Torsten > > > On 12/02/2014 08:28 PM, Jeff Hammond wrote: >> >> I created ticket 462 as a placeholder for whenever someone has time to >> work on this. >> >> MPI_GRAPH_CREATE does harm to users because they might use it. >> Somebody might naively think that distributed graphs are more >> complicated to use than replicated ones... >> >> Jeff >> >> On Tue, Dec 2, 2014 at 1:57 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: >>> >>> I am unfortunately too busy these days to drive it (a clean removal will >>> touch several places). It is ugly but it doesn't do any direct harm to >>> the >>> standard or its users. >>> >>> Best, >>> Torsten >>> >>> >>> On 01.12.2014 17:02, Jeff Hammond wrote: >>>> >>>> >>>> But you're not going to write that proposal? You seem more qualified >>>> than >>>> most... >>>> >>>> Jeff >>>> >>>> Sent from my iPhone >>>> >>>>> On Dec 1, 2014, at 7:51 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: >>>>> >>>>> Jeff, >>>>> >>>>> There is no such proposal yet (afaik), I'd support a carefully crafted >>>>> one and I think I wouldn't be alone. >>>>> >>>>> Thanks & Best, >>>>> Torsten >>>>> >>>>>> On 01.12.2014 06:41, Jeff Hammond wrote: >>>>>> Regarding http://lists.mpi-forum.org/mpiwg-coll/2014/03/0289.php, is >>>>>> there a proposal to deprecate MPI_GRAPH_CREATE? It seems that there >>>>>> are many issues with it and they are deliberately not getting fixed >>>>>> because a subset of the issues with it make it too horrible to fix. >>>>>> >>>>>> I find this situation distasteful as it leads to a poor user >>>>>> experience. Let's do the right thing and put MPI_GRAPH_CREATE out of >>>>>> our misery sooner rather than later. >>>>>> >>>>>> Thanks, >>>>>> >>>>>> Jeff >>>>> >>>>> >>>>> >>>>> -- >>>>> ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- >>>>> Torsten Hoefler | Assistant Professor >>>>> Dept. of Computer Science | ETH Zürich >>>>> Universitätsstrasse 6 | Zurich-8092, Switzerland >>>>> CAB E 64.1 | Phone: +41 76 309 79 29 >>>>> _______________________________________________ >>>>> mpiwg-coll mailing list >>>>> mpiwg-coll(a)lists.mpi-forum.org >>>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll >>>> >>>> >>>> _______________________________________________ >>>> mpiwg-coll mailing list >>>> mpiwg-coll(a)lists.mpi-forum.org >>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll >>>> >>> >>> -- >>> ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- >>> Torsten Hoefler | Assistant Professor >>> Dept. of Computer Science | ETH Zürich >>> Universitätsstrasse 6 | Zurich-8092, Switzerland >>> CAB E 64.1 | Phone: +41 76 309 79 29 >>> _______________________________________________ >>> mpiwg-coll mailing list >>> mpiwg-coll(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll >> >> >> >> > > -- > ### qreharg rug ebs fv crryF ------------- http://www.unixer.de/ ----- > > Torsten Hoefler | Assistant Professor > Dept. of Computer Science | ETH Zürich > Universitätsstrasse 6 | Zurich-8092, Switzerland > CAB E 64.1 | Phone: +41 76 309 79 29 > _______________________________________________ > mpiwg-coll mailing list > mpiwg-coll(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/

1 0

Re: [mpiwg-coll] deprecating MPI_GRAPH_CREATE?
by Torsten Hoefler 02 Dec '14

02 Dec '14

I consider this risk as low because it says it in the standard that it's non-scalable as well as in any recent documentation. Yes, some people will write bad codes. All the Best, Torsten On 12/02/2014 08:28 PM, Jeff Hammond wrote: > I created ticket 462 as a placeholder for whenever someone has time to > work on this. > > MPI_GRAPH_CREATE does harm to users because they might use it. > Somebody might naively think that distributed graphs are more > complicated to use than replicated ones... > > Jeff > > On Tue, Dec 2, 2014 at 1:57 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: >> I am unfortunately too busy these days to drive it (a clean removal will >> touch several places). It is ugly but it doesn't do any direct harm to the >> standard or its users. >> >> Best, >> Torsten >> >> >> On 01.12.2014 17:02, Jeff Hammond wrote: >>> >>> But you're not going to write that proposal? You seem more qualified than >>> most... >>> >>> Jeff >>> >>> Sent from my iPhone >>> >>>> On Dec 1, 2014, at 7:51 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: >>>> >>>> Jeff, >>>> >>>> There is no such proposal yet (afaik), I'd support a carefully crafted >>>> one and I think I wouldn't be alone. >>>> >>>> Thanks & Best, >>>> Torsten >>>> >>>>> On 01.12.2014 06:41, Jeff Hammond wrote: >>>>> Regarding http://lists.mpi-forum.org/mpiwg-coll/2014/03/0289.php, is >>>>> there a proposal to deprecate MPI_GRAPH_CREATE? It seems that there >>>>> are many issues with it and they are deliberately not getting fixed >>>>> because a subset of the issues with it make it too horrible to fix. >>>>> >>>>> I find this situation distasteful as it leads to a poor user >>>>> experience. Let's do the right thing and put MPI_GRAPH_CREATE out of >>>>> our misery sooner rather than later. >>>>> >>>>> Thanks, >>>>> >>>>> Jeff >>>> >>>> >>>> -- >>>> ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- >>>> Torsten Hoefler | Assistant Professor >>>> Dept. of Computer Science | ETH Zürich >>>> Universitätsstrasse 6 | Zurich-8092, Switzerland >>>> CAB E 64.1 | Phone: +41 76 309 79 29 >>>> _______________________________________________ >>>> mpiwg-coll mailing list >>>> mpiwg-coll(a)lists.mpi-forum.org >>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll >>> >>> _______________________________________________ >>> mpiwg-coll mailing list >>> mpiwg-coll(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll >>> >> >> -- >> ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- >> Torsten Hoefler | Assistant Professor >> Dept. of Computer Science | ETH Zürich >> Universitätsstrasse 6 | Zurich-8092, Switzerland >> CAB E 64.1 | Phone: +41 76 309 79 29 >> _______________________________________________ >> mpiwg-coll mailing list >> mpiwg-coll(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll > > > -- ### qreharg rug ebs fv crryF ------------- http://www.unixer.de/ ----- Torsten Hoefler | Assistant Professor Dept. of Computer Science | ETH Zürich Universitätsstrasse 6 | Zurich-8092, Switzerland CAB E 64.1 | Phone: +41 76 309 79 29

1 0

Re: [mpiwg-coll] deprecating MPI_GRAPH_CREATE?
by Jeff Hammond 02 Dec '14

02 Dec '14

I created ticket 462 as a placeholder for whenever someone has time to work on this. MPI_GRAPH_CREATE does harm to users because they might use it. Somebody might naively think that distributed graphs are more complicated to use than replicated ones... Jeff On Tue, Dec 2, 2014 at 1:57 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: > I am unfortunately too busy these days to drive it (a clean removal will > touch several places). It is ugly but it doesn't do any direct harm to the > standard or its users. > > Best, > Torsten > > > On 01.12.2014 17:02, Jeff Hammond wrote: >> >> But you're not going to write that proposal? You seem more qualified than >> most... >> >> Jeff >> >> Sent from my iPhone >> >>> On Dec 1, 2014, at 7:51 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: >>> >>> Jeff, >>> >>> There is no such proposal yet (afaik), I'd support a carefully crafted >>> one and I think I wouldn't be alone. >>> >>> Thanks & Best, >>> Torsten >>> >>>> On 01.12.2014 06:41, Jeff Hammond wrote: >>>> Regarding http://lists.mpi-forum.org/mpiwg-coll/2014/03/0289.php, is >>>> there a proposal to deprecate MPI_GRAPH_CREATE? It seems that there >>>> are many issues with it and they are deliberately not getting fixed >>>> because a subset of the issues with it make it too horrible to fix. >>>> >>>> I find this situation distasteful as it leads to a poor user >>>> experience. Let's do the right thing and put MPI_GRAPH_CREATE out of >>>> our misery sooner rather than later. >>>> >>>> Thanks, >>>> >>>> Jeff >>> >>> >>> -- >>> ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- >>> Torsten Hoefler | Assistant Professor >>> Dept. of Computer Science | ETH Zürich >>> Universitätsstrasse 6 | Zurich-8092, Switzerland >>> CAB E 64.1 | Phone: +41 76 309 79 29 >>> _______________________________________________ >>> mpiwg-coll mailing list >>> mpiwg-coll(a)lists.mpi-forum.org >>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll >> >> _______________________________________________ >> mpiwg-coll mailing list >> mpiwg-coll(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll >> > > -- > ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- > Torsten Hoefler | Assistant Professor > Dept. of Computer Science | ETH Zürich > Universitätsstrasse 6 | Zurich-8092, Switzerland > CAB E 64.1 | Phone: +41 76 309 79 29 > _______________________________________________ > mpiwg-coll mailing list > mpiwg-coll(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/

1 0

Re: [mpiwg-coll] deprecating MPI_GRAPH_CREATE?
by Torsten Hoefler 02 Dec '14

02 Dec '14

I am unfortunately too busy these days to drive it (a clean removal will touch several places). It is ugly but it doesn't do any direct harm to the standard or its users. Best, Torsten On 01.12.2014 17:02, Jeff Hammond wrote: > But you're not going to write that proposal? You seem more qualified than most... > > Jeff > > Sent from my iPhone > >> On Dec 1, 2014, at 7:51 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: >> >> Jeff, >> >> There is no such proposal yet (afaik), I'd support a carefully crafted one and I think I wouldn't be alone. >> >> Thanks & Best, >> Torsten >> >>> On 01.12.2014 06:41, Jeff Hammond wrote: >>> Regarding http://lists.mpi-forum.org/mpiwg-coll/2014/03/0289.php, is >>> there a proposal to deprecate MPI_GRAPH_CREATE? It seems that there >>> are many issues with it and they are deliberately not getting fixed >>> because a subset of the issues with it make it too horrible to fix. >>> >>> I find this situation distasteful as it leads to a poor user >>> experience. Let's do the right thing and put MPI_GRAPH_CREATE out of >>> our misery sooner rather than later. >>> >>> Thanks, >>> >>> Jeff >> >> -- >> ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- >> Torsten Hoefler | Assistant Professor >> Dept. of Computer Science | ETH Zürich >> Universitätsstrasse 6 | Zurich-8092, Switzerland >> CAB E 64.1 | Phone: +41 76 309 79 29 >> _______________________________________________ >> mpiwg-coll mailing list >> mpiwg-coll(a)lists.mpi-forum.org >> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll > _______________________________________________ > mpiwg-coll mailing list > mpiwg-coll(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll > -- ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- Torsten Hoefler | Assistant Professor Dept. of Computer Science | ETH Zürich Universitätsstrasse 6 | Zurich-8092, Switzerland CAB E 64.1 | Phone: +41 76 309 79 29

1 0

Re: [mpiwg-coll] deprecating MPI_GRAPH_CREATE?
by Jeff Hammond 01 Dec '14

01 Dec '14

But you're not going to write that proposal? You seem more qualified than most... Jeff Sent from my iPhone > On Dec 1, 2014, at 7:51 AM, Torsten Hoefler <htor(a)inf.ethz.ch> wrote: > > Jeff, > > There is no such proposal yet (afaik), I'd support a carefully crafted one and I think I wouldn't be alone. > > Thanks & Best, > Torsten > >> On 01.12.2014 06:41, Jeff Hammond wrote: >> Regarding http://lists.mpi-forum.org/mpiwg-coll/2014/03/0289.php, is >> there a proposal to deprecate MPI_GRAPH_CREATE? It seems that there >> are many issues with it and they are deliberately not getting fixed >> because a subset of the issues with it make it too horrible to fix. >> >> I find this situation distasteful as it leads to a poor user >> experience. Let's do the right thing and put MPI_GRAPH_CREATE out of >> our misery sooner rather than later. >> >> Thanks, >> >> Jeff > > -- > ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- > Torsten Hoefler | Assistant Professor > Dept. of Computer Science | ETH Zürich > Universitätsstrasse 6 | Zurich-8092, Switzerland > CAB E 64.1 | Phone: +41 76 309 79 29 > _______________________________________________ > mpiwg-coll mailing list > mpiwg-coll(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-coll

1 0

Re: [mpiwg-coll] deprecating MPI_GRAPH_CREATE?
by Torsten Hoefler 01 Dec '14

01 Dec '14

Jeff, There is no such proposal yet (afaik), I'd support a carefully crafted one and I think I wouldn't be alone. Thanks & Best, Torsten On 01.12.2014 06:41, Jeff Hammond wrote: > Regarding http://lists.mpi-forum.org/mpiwg-coll/2014/03/0289.php, is > there a proposal to deprecate MPI_GRAPH_CREATE? It seems that there > are many issues with it and they are deliberately not getting fixed > because a subset of the issues with it make it too horrible to fix. > > I find this situation distasteful as it leads to a poor user > experience. Let's do the right thing and put MPI_GRAPH_CREATE out of > our misery sooner rather than later. > > Thanks, > > Jeff > -- ### qreharg rug ebs fv crryF --- http://htor.inf.ethz.ch/ --- Torsten Hoefler | Assistant Professor Dept. of Computer Science | ETH Zï¿½rich Universitï¿½tsstrasse 6 | Zurich-8092, Switzerland CAB E 64.1 | Phone: +41 76 309 79 29

1 0

deprecating MPI_GRAPH_CREATE?
by Jeff Hammond 30 Nov '14

30 Nov '14

Regarding http://lists.mpi-forum.org/mpiwg-coll/2014/03/0289.php, is there a proposal to deprecate MPI_GRAPH_CREATE? It seems that there are many issues with it and they are deliberately not getting fixed because a subset of the issues with it make it too horrible to fix. I find this situation distasteful as it leads to a poor user experience. Let's do the right thing and put MPI_GRAPH_CREATE out of our misery sooner rather than later. Thanks, Jeff -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/

1 0