Matthijs' thoughts

• In MPI, how do the semantics of MPI compare to the semantics of grid computing? Is it, at all, sensible to span a single MPI job over multiple globally distributed resources? (cross-site MPI)
• Is there any relationship between efficiency of fabric usage efficiency, and the problem to be solved? (i.e. decomposition)
• In MPI, is there a solution to the endian/size differences when running on heterogeneous architectures?
• {C,S}hould grid jobs be crypto-signed before allowing them to use fabric? (to protect job integrity, and to encourage "due care" by programmers by having them feel more "responsible")
• What is the relationship between our work and Cactus Code? Can we do something to support it, integrate with it, or whatever?
• Would the GT4/webservices-based version of MPICH-G2, once released, be portable to (or downright compatible with) the gLite middleware? I.e., when, if ever, will the Globus Toolkit middleware disappear from the gLite distribution?

-- Main.mkoot - 21 Jun 2006