[Mpi3-tools] Next MPI-3 Tools Telcon 4/4/11
Martin Schulz
schulzm at llnl.gov
Sun Apr 3 04:00:45 CDT 2011
Hi all,
Following our regular schedule, our next MPI-3 Tools TelCon is
scheduled on 4/4 at the usual time 8am PDT, 11am PDT, 5pm
MSZ. Jeff has again setup a WebEx meeting:
https://cisco.webex.com/ciscosales/j.php?ED=162111667&UID=0&PW=NN2Q3ZjBlODUz&RT=MiMxMQ%3D%3D
Password: tools
At the MPI forum last week, we had the first reading for MPI_T
and generally, it went pretty well. However, there were a few
points that need to be addressed and that we need to go over
during the TelCon.
- The forum wanted a closer integration into the standard. Some
of this can be done with small language changes, but in some
cases we have similar concepts in MPI and MPI_T and we are
asked to make them the same if possible (or add more rationale
on why they have to be different):
- Error model / use of MPI error handlers
- Use of MPI types (must be usable before MPI_Init)
(Jeff is looking into the implementation complexity for Open MPI)
- How does the API deal with counter overflows? How is this
indicated to the user? Perhaps just an attribute saying that
a counter can overflow?
- How can we add to existing categories (e.g., when variables
are part of lazy loads)? Should we also allow shrinking?
- Should we add the option to atomically read multiple
variables?
- What text changes do we need to make to the overall MPI
standard?
Other edits that are missing
- Add additional performance variable class to show resource limits
- Make clear that indices can change between runs
- Add MPI_Info/MPI_Message to list of MPI objects
- Examples are still missing
In addition, it would be good if more people could read the text
for grammar and other small items. If you would like the TeX
file, but don't have access to the SVN, let me know.
Martin
________________________________________________________________________
Martin Schulz, schulzm at llnl.gov, http://people.llnl.gov/schulzm
CASC @ Lawrence Livermore National Laboratory, Livermore, USA
More information about the mpiwg-tools
mailing list