Speaker
Virginia Niculescu
(UBB)
Description
The hybrid nature of the new systems – distributed memory across nodes and shared memory within each node – poses challenges to application developers. The goal of this talk is to emphasize the differences between distributed and shared memory models, by presenting their fundamental operations and related developing strategies. Two traditional programming models are considered: MPI as a representative of the distributed memory model, and OpenMP that offers a very simple and in the same time efficient way to specify multithreading computation. We intend to present also how to create a basic hybrid program based on both, and how this approach could efficiently exploits the hardware facilities.
