PARALLELIZATION FOR MULTIPROCESSORS WITH MEMORY HIERARCHIES

Programming shared memory multiprocessors seems to be easier than developing programs for distributed memory systems. The reason is the existence of a global names space for parallel threads providing uniform access to all global data. This programming model seems to be inadequate for systems with a larger number of processors since memory hierarchies are integrated to eliminate the bottleneck of global memory access. Therefore programming these systems has to be done with respect to the distribution of data, thus to enforce the parallel processes exploiting locality of references. This paper describes an ongoing project in which we investigate the applicability of the distributed memory parallelization strategy for shared memory systems with memory hierarchies.