September 2008 - Mpi-22 - mailman3.cels.anl.gov

Tracking MPI-2.2 progress
by unknown＠example.com 08 Sep '08

08 Sep '08

Note that there are two ways to track what is happening in the MPI-2.2 effort: 1. The Trac we are using has an RSS feed of all activity: https://svn.mpi-forum.org/trac/mpi-forum-web/timeline?ticket=on&milestone=o… (you can tweak that URL a bit if you only want to see specific kinds of events, etc.) 2. You can add your e-mail address to the CC field of any ticket. Whenever changes are made to a ticket, the system will automatically send e-mail to the submitter, owner, and any e-mail address that is explicitly listed in the CC field. -- Jeff Squyres Cisco Systems

1 0

Moving proposals to wiki "tickets"
by unknown＠example.com 08 Sep '08

08 Sep '08

Short version: ============== All MPI-2.2 proposal authors are responsible for moving their current wiki proposals to tickets. See this URL for instructions: https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/mpi22/TicketWorkflow More details: ============= We have decided to move all the MPI-2.2 proposals to "tickets" (a la a tracking system, rather than continually editable web pages). This will allow us finer-grained searching capabilities, the ability to run reports (e.g., list tickets in a certain status), automatic e-mail updates when tickets are updated, fixed numbering, and so on. I have moved all of my proposals to tickets; the process is pretty straightforward. Most importantly, the wiki markup is exactly the same. Hence, it's mostly a copy-n-paste operation. All proposal authors are responsible for moving their proposals to tickets. Preserve any comments at the bottom of your current proposals as responses on the ticket. See this URL for information on how to create / edit tickets: https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/mpi22/TicketWorkflow -- Jeff Squyres Cisco Systems

1 0

Re: [Mpi-22] Slides from Dublin
by unknown＠example.com 08 Sep '08

08 Sep '08

Hi, Thanks. Here are the ABI final set and the collectives slides I showed. Best regards. Alexander -----Original Message----- From: mpi-22-bounces_at_[hidden] [mailto:mpi-22-bounces_at_[hidden]] On Behalf Of Jeff Squyres Sent: Friday, September 05, 2008 8:58 PM To: Main MPI Forum mailing list; MPI 2.2 Subject: [Mpi-22] Slides from Dublin As usual, if you presented slides in Dublin, please e-mail them to me and I will put them on the web site. Thanks. -- Jeff Squyres Cisco Systems _______________________________________________ mpi-22 mailing list mpi-22_at_[hidden] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22 --------------------------------------------------------------------- Intel GmbH Dornacher Strasse 1 85622 Feldkirchen/Muenchen Germany Sitz der Gesellschaft: Feldkirchen bei Muenchen Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer Registergericht: Muenchen HRB 47456 Ust.-IdNr. VAT Registration No.: DE129385895 Citibank Frankfurt (BLZ 502 109 00) 600119052 This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies. *

1 0

Re: [Mpi-22] Slides from Dublin
by unknown＠example.com 07 Sep '08

07 Sep '08

Here are my slides Bill On Sep 5, 2008, at 1:57 PM, Jeff Squyres wrote: > As usual, if you presented slides in Dublin, please e-mail them to me > and I will put them on the web site. > > Thanks. > > -- > Jeff Squyres > Cisco Systems > > _______________________________________________ > mpi-22 mailing list > mpi-22_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22 William Gropp Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign *

1 0

New proposal: Concurrent MPI_Init() and MPI_Init_thread()
by unknown＠example.com 07 Sep '08

07 Sep '08

Hi all, I've added a new proposal on the 2.2 wiki. Here's the gist of it: The MPI 2.1 standard does not explicitly mention whether it is permitted for one process in an application to initialize with a different thread-level than another. This needs to be clarified. Link to the proposal: https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/mpi22/concurrent_init_ini… -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

MPI 2.2 Summary
by unknown＠example.com 07 Sep '08

07 Sep '08

I'd like to thank everyone for a very productive MPI 2.2 session and especially Jeff Squyres for keeping records and updating the wiki with the results in real time - that was an enormous help. I encourage everyone to read the proposals very carefully and add comments directly to the wiki entries. We need more input from users of MPI as well; please involve any you know in these discussions. Thanks! Bill William Gropp Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign

1 0

Re: [Mpi-22] Dynamic thread levels
by unknown＠example.com 05 Sep '08

05 Sep '08

Hi all, Based on feedback from a number of folks (thanks!), I have updated the wiki with some more information, clarifications and examples. Please take a look at it. Thanks. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Slides from Dublin
by unknown＠example.com 05 Sep '08

05 Sep '08

As usual, if you presented slides in Dublin, please e-mail them to me and I will put them on the web site. Thanks. -- Jeff Squyres Cisco Systems

1 0

Re: [Mpi-22] MPI::Grequest::Start proposal
by unknown＠example.com 04 Sep '08

04 Sep '08

FWIW, I checked this out with Doug Gregor (my C++-goto-guy) and he basically said "yep, that's right." On Sep 4, 2008, at 9:02 AM, Erez Haba wrote: > I think that this is okay; when we discussed it, I thought it was a > different fix. The proposal is just fine. > > -----Original Message----- > From: mpi-22-bounces_at_[hidden] [mailto:mpi-22-bounces_at_[hidden] > ] On Behalf Of Jeff Squyres > Sent: Thursday, September 04, 2008 7:26 AM > To: MPI 2.2 > Subject: [Mpi-22] MPI::Grequest::Start proposal > > Regarding https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/mpi22/GrequestStartFnPtrA… > : > > I have confirmed in Open MPI that if you change this: > > static Grequest Start(Query_function, Free_function, > Cancel_function, void *); > > to > > static Grequest Start(Query_function *, Free_function *, > Cancel_function *, void *); > > Recompile and reinstall, both the implementation and same C++ user > code compiles without warning/error. > > Erez raised the point that this would change the type of any > implementation and user-declared variables that hold the function > pointer. It does not; such a variable type must always be > (Free_function*); it's only the type that is passed through a function > argument that can be either (Free_function) or (Free_function*). > Someone smarter than me in C++ can explain why. :-) > > Specifically, the following compiles and runs without warning/error: > > ----- > #include <stdio.h> > typedef int Free_function(void *); > > int my_function(void*) { > printf("In my_function\n"); > return 0; > } > > void foo1(Free_function ptr) { > Free_function *save = ptr; > save(0); > } > void foo2(Free_function* ptr) { > Free_function *save = ptr; > save(0); > } > int main (int argc, char*argv[]) { > foo1(my_function); > foo2(my_function); > return 0; > } > ----- > > I tried 4 different C++ compilers (gnu, intel, pgi, pathscale): > > ----- > [23:19] svbu-mpi:~/tmp % g++ foo.cc -o foo && ./foo > In my_function > In my_function > [23:21] svbu-mpi:~/tmp % icpc foo.cc -o foo && ./foo > In my_function > In my_function > [23:21] svbu-mpi:~/tmp % pgCC foo.cc -o foo && ./foo > In my_function > In my_function > [23:22] svbu-mpi:~/tmp % pathCC foo.cc -o foo && ./foo > In my_function > In my_function > [23:22] svbu-mpi:~/tmp % > ------ > > So I think the proposal stands as it is written. > > -- > Jeff Squyres > Cisco Systems > > _______________________________________________ > mpi-22 mailing list > mpi-22_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22 > > > _______________________________________________ > mpi-22 mailing list > mpi-22_at_[hidden] > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-22 -- Jeff Squyres Cisco Systems

1 0

Re: [Mpi-22] Dynamic thread levels
by unknown＠example.com 04 Sep '08

04 Sep '08

> I've implemented this in MPICH2 and placed a copy here: > https://svn.mcs.anl.gov/repos/mpi/mpich2/branches/dev/dyn_thread_level. > Based on this implementation, I did some measurements using a single > threaded application, but using MPI_Init_thread() vs. MPI_Init(). > > Experiment 1: Each process sends to MPI_PROC_NULL (emulating an > infinitely fast network) > > MPI_Init -- 29.94 million messages per second > MPI_Init_thread -- 15.86 million messages per second > Difference -- almost 2X > > Experiment 2: Two processes communicating over TCP > > MPI_Init -- 3.947 million messages per second > MPI_Init_thread -- 3.231 million messages per second > Difference -- about 20% Minor clarification -- with the implementation pointed out above (which adds a new MPIX_Set_thread_level() call), MPI_Init_thread() + MPIX_Set_thread_level() to serial has the same performance as regular single-threaded MPI_Init(). -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0