One topic for the first meeting would be coming up with a clear definition/description of "Fault Tolerance" in the context of MPI. Obviously, we can refine it over email prior to teh confcall.
More specifically, I think we need to break the overall task of application fault tolerance into application responsibilities, MPI responsibilities and third party responsibilities. For example, an MPI responsibility may be to maintain stability in the face of non-catastrophic node and link failures (i.e. most nodes are alive and no network partition). An application responsibility may be to survive the failure of one or more nodes through some application-specific technique or through checkpointing. A third-party responsibility may be checkpointing, which may be included as part of the application or embedded into MPI but will probably not be specified as part of the MPI spec. Greg Bronevetsky Post-Doctoral Researcher 1028 Building 451 Lawrence Livermore National Lab (925) 424-5756 [email protected]