Skip to content

2018 05 23

Jump to bottom
Wesley Bland edited this page May 23, 2018 · 1 revision

FTWG Meeting Notes 2018-05-23

ULFM

  • Aurelien read through the proposals about the definitions of MPI_INIT and MPI_FINALIZE in the event of an error.
    • Feedback was that it was still referring to fault tolerance too much and should be more focused on error handling.
      • We should make sure that we're not attaching particular implementation details to MPI_INIT, such as that MPI_COMM_WORLD will be set up when MPI_INIT is done.
    • The was also conversation about how we want to do fail-fast for MPI_COMM_WORLD.
      • If it's important to the application to know that MPI_COMM_WORLD is set up correctly, it should do an MPI_BARRIER after MPI_INIT.
Clone this wiki locally