并行稀疏矩阵与向量乘的负载平衡和通信优化

Load balancing and communication optimization for parallel sparse matrix-vector multiplication

本文考虑了在工作站机群上实现大型稀疏矩阵和向量乘的负载平衡。提出了一个快速负载平衡和有效的消息传递技术相结合的方法,来缓解计算和节点间通信。该方法的性能在工作站机群上进行测试,可获得良好结果;并且,通过I/O延迟隐藏和整体负载平衡使I/O开销能有效地分摊。

The load-balanced multiplication of a large sparse matrices with vectors on workstation cluster is considered in this paper. A method that combines fast load balancing with efficient message-passing techniques to alleviate computation and inter-node communication is presented. The performance of the method is evaluated on workstation cluster and a good result is obtained. Moreover, it is also shown that I/O overhead can be efficiently amortized through I/O latency hiding and overall load balancing.