Research on optimization of large-scale sparse matrix communication in parallel application
大规模稀疏矩阵在并行应用中的通信优化研究*

This paper presented a method of compressed communication for sparse matrix in order to improve the performance of communication.obtained a formula for calculating the lower bound of sparsity by quantitative analysis of the relationship between sparsity, network bandwidth and frequency of CPU. In addition,derived a function between sparsity and communication performance from analysis of the algorithm efficiency on different sparsities. Based on the application of parallel petroleum reservoir simulation,implemented an algorithm for compressed communication.The experiments show that the performance of sparse matrix communication can be improved significantly.