In this paper, we proposed a unified framework and try to address the optimal block size selection problem for paral-lel blocked LU factorization based on ScaLAPACK package since it uses block cyclic data distribution fashion, block size plays important role in determining the final perfor-mance. Through the analysis with our proposed framework and experiments on small scale system configuration, we found that among all these factors, load balance and lo-cal block size selection play key roles in determining the optimal block size on SR2201(pseudo-vector based MPP machine). The optimal block size is determined by the pro-cessor grid shape and problem size. Bssed on this observa-tion, an optimal block size prediction formula with proces-sor grid shape and problem size as parameters was given that can match with the experimental results well. The ap-plication of our framework on scalar based parallel ma-chines and on other applications program wound be the fu-ture work.
展开▼