Abstract: This talk will explore some of the obstacles (and promises) of scaling MPI, the Message-Passing Interface, to massively parallel systems with millions of processors.  The exploration will consider both the MPI interface itself, discussing improvements and changes for better supporting massive scaling that are being addressed in the currently active MPI Forum, as well as new issues related to the efficient implementation of MPI on such systems. Some of the issues are also relevant for implementing MPI on emerging, memory- and bandwidth constrained multi-core systems.

The talk is based on joint work with Pavan Balaji, Darius Buntinas, David Goodell, William Gropp, Sameer Kumar, Ewing Lusk, and Rajeev Thakur.

Homepage of Dr. Jesper Larsson Traeff >>more

NEC Europe >>more