INTERTWinE's Dan Holmes reports from MPI Forum meeting

05 Oct 2017
It’s official - MPI is moving too slowly and we should do something about it!

The MPI Forum has decided that now is the time to accelerate its work on defining the next version of the MPI Standard. All active working groups will move to weekly meetings instead of meeting every two weeks and there are proposals to change the formal procedures so that technical proposals can progress faster.

There is now a greater sense of urgency around key proposals such as:

  • How can we encourage performance planning in MPI? We are very close to final acceptance for new persistent collective operations but can we also move forward with other persistent operations, such as I/O and Streams?
  • How should MPI interact with threads and tasks? Is MPI Endpoints the right way to do that or do we need something else?
  • How should MPI manage dynamic resources? How can we continue to extend MPI Sessions to deal with adaptive applications?
  • How should MPI handle faults? Is process fail-stop the best fault model for MPI? Should we go ahead with ULFM or hold out for something better?

The INTERTWinE team is leading some of these areas as well as tracking and contributing to all of the others.

These are exciting times for MPI - there are some major changes coming for MPI-4. You can get involved and have your say by joining the weekly teleconferences for any working group that interests you, or by attending the face-to-face meetings. For more information, contact us here at INTERTWinE: info [at] intertwine-project.eu.

Last updated: 05 Oct 2017 at 11:31