Improve the way of assigning blocks to multiple cores with MPI, when different cores may have different performance.