This is actually rebuild the wheel project, practice on MPI and obtain better understand on sparse matrix operators The geom structure is 2D, the cpu topology is 1D or 2D, while the performance of 2D is not that good as I thought A great project online can be found: HPCG