How might MPI process location aff ect performance of a 3 D code with nearest-neighbor communication? On a s trong s caling pa rallel r un, t he p oint w here t he w orking set of an application fi...

How might MPI process location aff ect performance of a 3 D

code with nearest-neighbor communication?

On a s trong s caling pa rallel r un, t he p oint w here t he w orking set of an application fi ts into cache exhibits a su perlinear

speedup. Can this be achieved with an OpenMP? What are the

requirements on an OpenMP loop that might exhibit such a

superlinear speedup.