并行高斯消去法在云计算平台上的研究.PDFVIP

网络出版时间：2014-02-11 16:13 网络出版地址：/kcms/detail/61.1450.TP1613.032.html 并行高斯消去法在云计算平台上的研究 潘晓辉 (上海政法学院，现代教育技术中心，上海 201701) 摘 要：为了解决串行部分选主元的高斯消去算法不能充分利用多核处理器的问题，提出并 实现了并行多线程的部分选主元的高斯消去算法，并将整个算法进行了分析和优化，使数据 的存储布局和算法的访存模式匹配，从而大幅提高了程序的性能。通过对本地Linux服务器 以及美国亚马逊EC2云的多种平台上的实验结果的比较和分析，确定了部分选主元的高斯消 去算法受缓存影响较大，所以在CPU和内存/缓存配置较为均衡的平台上运行性能最好。本文 展现了一种高效率、扩展性好的多线程并行部分选主元的高斯消去算法以及将一般性串行算 法进行并行化和优化的方法。 关键词: 部分选主元;高斯消去算法;多线程;缓存;亚马逊云 中图分类号: TP391.9 文献标志码 :A TheStudyoftheParallelGaussianEliminationAlgorithmontheCloudPlatform PANXiaohui （1ModernEducationTechnologyCenter,ShanghaiUniversityofPoliticalScienceandLaw, Shanghai201701，China) Abstract: To solve the problem that the sequential Gaussian Elimination with partial pivoting algorithmcannotutilize allthe coresofamulti-core CPU,anovelparallelmulti-threaded Gaussian Elimination with partial pivoting algorithm is proposed and implemented. The whole algorithm is then analyzed and optimized suchthat the storage layout of datamatches the accesspattern ofthe algorithm;thereforetheapplicationperformanceisgreatlyimproved.Theapplicationisexecutedon multiple computer platforms including a Linux server and several platforms in Amazon Elastic Compute Cloud.The comparison and analysis ofthe experimentresults showthatthe performance ofthemulti-threadedGaussianEliminationwithpartialpivotingalgorithmextremelydependsonthe cache system. The best performance is achieved on a platform with proportionate CPU and memory/cache hardware. Thus ahigh-efficiency and scalablemulti-threaded Gaussian Elimina