Abstract: The new features of multi-core make the memory hierarchy of multi-core clusters more complex, and also add the optimization space for MPI programs. We tested the communication performance of three different multi-core clus- ters, and evaluated some general optimization technologies, such as hybrid MPI/OpenMP, tuning MPI runtime parame- ters and optimization of MPI process placement in Intel and AMD multi core cluster. I}he experiments result and opti- mization performances were also analyzed.

Key words: Multi-core cluster, Memory hierarchy, MPI programs optimization, Hybrid MPI/OpenMP, MPI runtime parameters, MPI process placement