March 2010 - mpiwg-hybridpm - mailman3.cels.anl.gov

Re: [Mpi3-hybridpm] Reminder for telecon: 03/30
by Snir, Marc 31 Mar '10

31 Mar '10

I skipped because of Passsover -- sorry. On Mar 29, 2010, at 12:32 AM, Pavan Balaji wrote: > All, > > This is the reminder for the telecon on Tuesday (03/30). Telecon > information is available on the wiki. > > The primary agenda item is still the question raised at the forum with > respect to what the recommended return value of MPI_MAX_ENDPOINTS would > be. We discussed some solutions last time; we'll continue on that. > > Doug: did you make any changes to your proposal with respect to > decoupling it from endpoints? If yes, it'll be good to discuss that as well. > > Thanks, > > -- Pavan > > -- > Pavan Balaji > http://www.mcs.anl.gov/~balaji > _______________________________________________ > Mpi3-hybridpm mailing list > Mpi3-hybridpm(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm

1 0

Re: [Mpi3-hybridpm] Reminder for telecon: 03/30
by Pavan Balaji 30 Mar '10

30 Mar '10

We had a short call. Here are the meeting notes: https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/MPI3Hybrid/notes-2010-03-… -- Pavan Pavan Balaji wrote: > All, > > This is the reminder for the telecon on Tuesday (03/30). Telecon > information is available on the wiki. > > The primary agenda item is still the question raised at the forum with > respect to what the recommended return value of MPI_MAX_ENDPOINTS would > be. We discussed some solutions last time; we'll continue on that. > > Doug: did you make any changes to your proposal with respect to > decoupling it from endpoints? If yes, it'll be good to discuss that as well. > > Thanks, > > -- Pavan > -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Re: [Mpi3-hybridpm] Reminder for telecon: 03/30
by Pavan Balaji 29 Mar '10

29 Mar '10

Douglas Miller wrote: > It is the leave that blocks, that is where the threads actually perform > work. Between join and leave, if a thread makes calls to MPI_Wait (et al.) > - presumably for it's own communications - then it will also perform any > other work assigned to it at that time. Ah, yes. I remember now. Sorry for the unnecessary exchange of emails. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Re: [Mpi3-hybridpm] Reminder for telecon: 03/30
by Douglas Miller 29 Mar '10

29 Mar '10

It is the leave that blocks, that is where the threads actually perform work. Between join and leave, if a thread makes calls to MPI_Wait (et al.) - presumably for it's own communications - then it will also perform any other work assigned to it at that time. _______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A410 dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM Pavan Balaji <balaji(a)mcs.anl.g ov> To Sent by: mpi3-hybridpm(a)lists.mpi-forum.org mpi3-hybridpm-bou cc nces(a)lists.mpi-fo rum.org Subject Re: [Mpi3-hybridpm] Reminder for telecon: 03/30 03/29/2010 09:35 AM Please respond to mpi3-hybridpm@lis ts.mpi-forum.org Hi Doug, Douglas Miller wrote: > The example is as I intended. The usage I've been promoting is that all > team members do the join and leave. This is the only way to get all threads > participating. Sorry if I'm forgetting something we already discussed -- in your example, are the t != 0 threads blocking inside MPI_Endpoint_join()? My guess is no, since the t == 0 thread which needs to make the Allgather call needs to come out. Then how will MPI use these threads (they need to be under MPI's control to be used). -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji _______________________________________________ Mpi3-hybridpm mailing list Mpi3-hybridpm(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm

1 0

Re: [Mpi3-hybridpm] Reminder for telecon: 03/30
by Pavan Balaji 29 Mar '10

29 Mar '10

Hi Doug, Douglas Miller wrote: > The example is as I intended. The usage I've been promoting is that all > team members do the join and leave. This is the only way to get all threads > participating. Sorry if I'm forgetting something we already discussed -- in your example, are the t != 0 threads blocking inside MPI_Endpoint_join()? My guess is no, since the t == 0 thread which needs to make the Allgather call needs to come out. Then how will MPI use these threads (they need to be under MPI's control to be used). -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Re: [Mpi3-hybridpm] Reminder for telecon: 03/30
by Douglas Miller 29 Mar '10

29 Mar '10

The example is as I intended. The usage I've been promoting is that all team members do the join and leave. This is the only way to get all threads participating. I had originally not had the team create/destroy, but our discussions last time led to what seemed to be agreement on using a create/destroy semantic for teams. We can certainly revisit this during our discussions. thanks. _______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A410 dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM Pavan Balaji <balaji(a)mcs.anl.g ov> To Sent by: mpi3-hybridpm(a)lists.mpi-forum.org mpi3-hybridpm-bou cc nces(a)lists.mpi-fo rum.org Subject Re: [Mpi3-hybridpm] Reminder for telecon: 03/30 03/29/2010 08:58 AM Please respond to mpi3-hybridpm@lis ts.mpi-forum.org Douglas Miller wrote: > I do have an updated proposal in Ticket #217 > (https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/217) and a link to > this from the Hybrid Programming wiki. I'll need some input to get this > into a proper format for proposal, but hopefully the necessary content is > there. Thanks. The example on page 3 doesn't seem to match the description you provided a few calls ago. This was my understanding: MPI_Helper_team_create(); if (thread_id != master_thread) MPI_Helper_join(); if (thread_id == master_thread) { MPI_Allgather(...); MPI_Helper_leave(); /* this releases all the threads */ } MPI_Helper_team_destroy(); Did I miss something here? Secondary point -- we might consider passing the number of threads as a parameter to Helper_join, rather than create a new call MPI_Helper_team_create/destroy. Do you they serve any other purpose than just telling how many threads will be joining in? > I'll have to be dropping off the conference call at 11:30, so if possible > can we discuss this first? Sure, we can do that. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji _______________________________________________ Mpi3-hybridpm mailing list Mpi3-hybridpm(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm

1 0

Re: [Mpi3-hybridpm] Reminder for telecon: 03/30
by Pavan Balaji 29 Mar '10

29 Mar '10

Douglas Miller wrote: > I do have an updated proposal in Ticket #217 > (https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/217) and a link to > this from the Hybrid Programming wiki. I'll need some input to get this > into a proper format for proposal, but hopefully the necessary content is > there. Thanks. The example on page 3 doesn't seem to match the description you provided a few calls ago. This was my understanding: MPI_Helper_team_create(); if (thread_id != master_thread) MPI_Helper_join(); if (thread_id == master_thread) { MPI_Allgather(...); MPI_Helper_leave(); /* this releases all the threads */ } MPI_Helper_team_destroy(); Did I miss something here? Secondary point -- we might consider passing the number of threads as a parameter to Helper_join, rather than create a new call MPI_Helper_team_create/destroy. Do you they serve any other purpose than just telling how many threads will be joining in? > I'll have to be dropping off the conference call at 11:30, so if possible > can we discuss this first? Sure, we can do that. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Re: [Mpi3-hybridpm] Reminder for telecon: 03/30
by Douglas Miller 29 Mar '10

29 Mar '10

I do have an updated proposal in Ticket #217 (https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/217) and a link to this from the Hybrid Programming wiki. I'll need some input to get this into a proper format for proposal, but hopefully the necessary content is there. I'll have to be dropping off the conference call at 11:30, so if possible can we discuss this first? thanks, _______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A410 dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM Pavan Balaji <balaji(a)mcs.anl.g ov> To Sent by: mpi3-hybridpm(a)lists.mpi-forum.org mpi3-hybridpm-bou cc nces(a)lists.mpi-fo rum.org Subject [Mpi3-hybridpm] Reminder for telecon: 03/30 03/29/2010 12:32 AM Please respond to mpi3-hybridpm@lis ts.mpi-forum.org All, This is the reminder for the telecon on Tuesday (03/30). Telecon information is available on the wiki. The primary agenda item is still the question raised at the forum with respect to what the recommended return value of MPI_MAX_ENDPOINTS would be. We discussed some solutions last time; we'll continue on that. Doug: did you make any changes to your proposal with respect to decoupling it from endpoints? If yes, it'll be good to discuss that as well. Thanks, -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji _______________________________________________ Mpi3-hybridpm mailing list Mpi3-hybridpm(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm

1 0

[Mpi3-hybridpm] Reminder for telecon: 03/30
by Pavan Balaji 29 Mar '10

29 Mar '10

All, This is the reminder for the telecon on Tuesday (03/30). Telecon information is available on the wiki. The primary agenda item is still the question raised at the forum with respect to what the recommended return value of MPI_MAX_ENDPOINTS would be. We discussed some solutions last time; we'll continue on that. Doug: did you make any changes to your proposal with respect to decoupling it from endpoints? If yes, it'll be good to discuss that as well. Thanks, -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Re: [Mpi3-hybridpm] Agenda for telecon: 03/16
by Pavan Balaji 16 Mar '10

16 Mar '10

Here are the meeting minutes for today's telecon: https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/MPI3Hybrid/notes-2010-03-… I didn't quite catch everyone who attended the telecon, so if I missed someone, please let me know. Thanks, -- Pavan On 03/14/2010 10:13 PM, Pavan Balaji wrote: > Folks, > > For the telecon on 03/16, we'll mainly focus on the feedback we got from > the Forum to see how we can address it. > > Telecon information is available on the wiki. > > Regards, > > -- Pavan > -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0