基于多核并行的海量数据序列模式挖掘  

Parallel massive mining of sequential patterns based on multi-core processors

作　　者：俞东进[1] 郑苏杭[1] 李万清[1] 

机构地区：[1]杭州电子科技大学计算机学院,杭州310018

出　　处：《计算机应用研究》2012年第2期478-481,521,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(60903053);浙江省重大科技计划资助项目(2008C11099-1)

摘　　要：为了在多核处理器上充分利用多核资源以提升挖掘性能,提出了一种动态与静态任务分配机制相结合的基于多核的并行序列模式挖掘算法。该算法采用数据并行与任务并行相结合的策略,在各处理器核生成局部序列模式后,再与其他处理器核协同,以最终获得所有的全局序列模式。算法通过并行局部归约技术消除了局部序列的重复生成与计算,并可结合静态与动态任务分配机制解决处理器的负载不均衡问题。理论分析和实验都证实了该算法可有效利用多核计算平台及多核体系结构优势,具有较高的运行效率和加速比。To fully utilize the multi-core resources on multi-core processors to improve mining performance,this paper presented a novel algorithm of mining parallel sequential patterns based on the multi-core processors.It combined the data parallelism and task parallelism,with global sequential patterns obtained by combining local patterns discovered in different processor cores.Through local parallel reduction,it eliminated the repetitive patterns and reduced computational effort.Besides,it achieved the workload balancing by static and dynamic task distribution mechanisms.Both theoretical analysis and practical experiments show that the algorithm takes good advantage of multi-core computing platform,having higher operating efficiency and speedup.

关 键 词：并行 多核 序列模式 海量数据挖掘 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]