I am wondering how the ideas scetched in the attachment which was just distributed map to any OpenMP coding style. Spawning helper threads for a "random period of time"? How would this work in OpenMP? OpenMP is not specified to work well with additional native threads (yet), allthough some current implementations do to some degree. Dieter ----- Original Message ----- From: Douglas Miller <[email protected]> Date: Tuesday, February 16, 2010 2:51 pm Subject: Re: [Mpi3-hybridpm] Reminder for telecon tomorrow To: [email protected]

I am definitely thinking about OpenMP and how best to use it, but of course we can't depend on it being used in all cases so we have to have solutions that work with pthreads or other mechanisms for parallelism.

_______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A410 [email protected] Douglas Miller/Rochester/IBM

"an Mey, Dieter"

<[email protected]

chen.de> To Sent by: "[email protected]" mpi3-hybridpm-bou <[email protected]> [email protected] cc rum.org

Subject Re: [Mpi3-hybridpm] Reminder for 02/16/2010 05:08 telecon tomorrow

AM

Please respond to

mpi3-hybridpm@lis

ts.mpi-forum.org

Dear all,

I subscribed to this list only recently in order to get an impression on how MPI is evolving concerning hybrid parallelization. I am participating in the OpenMP Language Committee and I am wondering in how far OpenMP is of any concern.

Best regards, Dieter

-- Dipl.-Math. Dieter an Mey, HPC Team Lead RWTH Aachen University, Center for Computing and Communication Rechen- und Kommunikationszentrum der RWTH Aachen Seffenter Weg 23, D 52074 Aachen (Germany) Phone: + 49 241 80 24377 - Fax/UMS: + 49 241 80 624377 http://www.rz.rwth-aachen.de

...

-----Original Message----- From: [email protected] [mailto:mpi3-hybridpm- [email protected]] On Behalf Of Pavan Balaji Sent: Tuesday, February 16, 2010 1:25 AM To: [email protected] Subject: Re: [Mpi3-hybridpm] Reminder for telecon tomorrow

On 02/15/2010 02:51 PM, Douglas Miller wrote:

...

Here is the document for starting a discussion on additional parallelism models.

Thanks Doug. I think the biggest issue here is the lack of a concrete API and semantics. This (at least for me) was the cause of the most confusion. That is, can a user provide a helper thread to the MPI implementation for a random period of time? How does the thread detach (does some other thread detach or does the MPI implementation "let go" of the thread when it's done)?

If you can add some notes about these, that'll be great. Either ways, we can discuss these issues during tomorrow's call.

-- Pavan

-- Pavan Balaji http://www.mcs.anl.gov/~balaji _______________________________________________ Mpi3-hybridpm mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm

---

Mpi3-hybridpm mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm

_______________________________________________ Mpi3-hybridpm mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm