SCRS并行算法的全局通信策略

发布时间：2018-05-29 10:39

  本文选题：大型稀疏线性方程组 + SCRS法　； 参考：《福建师范大学学报(自然科学版)》2017年04期

【摘要】：基于Gu等人提出的全局通信策略,改进平滑共轭残量平方法 (SCRS),提出了改进的适合于分布式并行计算环境的SCRS算法(ISCRS).通过改变算法的计算顺序,使得ISCRS算法相对于SCRS算法将3个全局同步点减少为1个.每次迭代的所有的内积是独立的,且内积所需的通信时间能和计算有效重叠.理论分析和数值实验指出ISCRS比SCRS有更好的并行性和可扩展性.
[Abstract]:Based on the global communication strategy proposed by Gu et al, an improved smooth conjugate residual balance method is proposed, and an improved SCRS algorithm for distributed parallel computing environment is proposed. By changing the calculation order of the algorithm, the ISCRS algorithm can reduce the three global synchronization points to one compared with the SCRS algorithm. All the inner products of each iteration are independent, and the communication time required for the inner product overlaps with the computation effectively. Theoretical analysis and numerical experiments show that ISCRS has better parallelism and scalability than SCRS.
【作者单位】： 福建师范大学数学与计算机科学学院;福建师范大学福清分校电子与信息工程学院;
【基金】：福建省自然科学基金资助项目(2014J01006)
【分类号】：O241.6
，

本文编号：1950573