I'll also point out that if there's an MPI implementation where an MPI process is a thread in an OS process, and there can be multiple MPI processes in an OS process, then we have no way to represent that in the current MPIR Process Acquisition spec. Currently, MPIR assumes "MPI process" == "OS process". Cheers, John D. Martin Schulz wrote:

On Sep 12, 2013, at 6:25 PM, Jeff Squyres (jsquyres) <[email protected]> wrote:

...

On Sep 12, 2013, at 5:35 PM, Martin Schulz <[email protected]> wrote:

...

Also, there were questions on the definition of a "unique image" - at the end people seemed to be OK with it, but would like to see a cleaner definition for the next forward looking version of MQD. In particular, it should cover the case where multiple MPI processes are in one OS process. I'm not too concerned about this particular piece of feedback:

1. It came from Torsten who had never read the proposal before. 2. If the document doesn't handle the MPI-processes-as-threads concept, then so be it -- this is a documentation of current practice. We welcome their proposal for v2.0 of the document.

I agree (in case this didn't come across) - it should be OK for the current version, but will be a point of discussion once/if we work on the next generation of the next document.

Martin

...

-- Jeff Squyres [email protected] For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

_______________________________________________ mpiwg-tools mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-tools

_______________________________________________ mpiwg-tools mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-tools