Jeff,

While we can debate the sanity of globally-scoped data at exascale, there will always be science which is not amenable to simple domain decomposition and thus requires some type of global view. Whether the data structures are simple ala GA or complex ala Madness, RMI is the only sensible ways to implement these involve RMI.

just as a side-note: what you call RMI is what I'd call Active Message and I fully agree that this is one key technology to reach good scalability for a class of applications (I was mostly thinking of irregular applications but your point is an excellent addition). However, if your problem is static (enough), you could establish channels upfront (e.g., sparse matrix vector in Krylov subspace methods). Best, Torsten -- bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ ----- Torsten Hoefler | Research Associate Blue Waters Directorate | University of Illinois 1205 W Clark Street | Urbana, IL, 61801 NCSA Building | +01 (217) 244-7736