Hi all, At 06:00 PM 6/10/2009, Greg Bronevetsky wrote:
For the next telecon I propose that we address the following topics. - The piggybacking proposal was considered by the FT group several months ago but then got lost in a limbo between the FT group and the tools group. We need to discuss this proposal and the issues behind it so that we can figure out how to make progress on it.
I agree with Greg - it would be great if we could pick this up again and try to come up with a single, common proposal. There is interest in this from multiple WGs (which is probably part of the problem of why nobody has pushed this forward) and if the FT group could continue to "host" this, that would be great. It would also be good to cross-post the TelCon announcement to the tools and AM WGs (for the latter, Torsten has voiced interest). Thanks, Martin
- During the face-to-face discussion we began talking about the desired error notification semantics for collectives. The question is: are MPI implementations allowed to provide no error notification as part of their collectives, relying on calls to MPI_Validate for this functionality or do we expect them to report all relevant errors. In particular, is MPI_Barrier responsible for reporting failures of any members of the communicator or is it allowed to mis-behave in the event of failures and for example, return successfully even if one process failed before calling MPI_Barrier? If so, how do we bound the limit of such mis-behaviors?
Greg Bronevetsky Post-Doctoral Researcher Lawrence Livermore National Lab (925) 424-5756 [email protected] http:// greg.bronevetsky.com _______________________________________________ mpi3-ft mailing list [email protected] http:// lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-ft
_______________________________________________________________________ Martin Schulz, [email protected], http://people.llnl.gov/schulz6 CASC @ Lawrence Livermore National Laboratory, Livermore, USA