基于MPI的海量数据线性拟合并行算法研究  

Research on Parallel Algorithm of Mass Data Fitting Based on MPI

作　　者：冯小丹[1] 梁立[1] 袁凌云[1] 高丽金[1] 王文元 

机构地区：[1]云南师范大学计算机科学与信息技术学院,云南昆明650092 [2]红云红河烟草(集团)技术中心生理生化研究室,云南昆明650202

出　　处：《南京师大学报（自然科学版）》2010年第4期157-160,共4页Journal of Nanjing Normal University(Natural Science Edition)

基　　金：云南省社会发展基础研究(面上项目)(2008CD113);云南省教育厅基金(08Y0136)

摘　　要：给出了线性模型数据拟合的并行算法描述,分析了该并行算法的任务分配、通信开销、负载平衡等问题;通过多项式拟合在PC集群环境下的实现,比较了串行多项式拟合与并行多项式拟合运算时耗的差别,分析并行算法的性能,从而探索建立集群环境对于解决海量数据多项式拟合次数范围大、运算速度过慢等问题的可行性.The linear model fitting of parallel algorithm is described, the distribution of tasks, communication costs, load balancing and other issues of the parallel algorithm are analyzed. Via the realization of polynomial parallel fitting in the PC cluster environment, the running time difference between the serial and parallel polynomial fitting is compared and the performance of the parallel algorithm is discussed. The experimental results confirm the feasibility of establishing a cluster environment for solving the problems exist in large data polynomial fitting, such as the wide range and the slow calculation speed.

关 键 词：MPI 集群系统 并行算法 数据拟合 线性 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]