面向云环境的集群资源模糊聚类划分算法的优化  

Optimization of Cluster Resource Fuzzy Clustering Partition Algorithm for Cloud Computing

作　　者：董世龙[1] 陈宁江[1] 谭瑛[1] 何子龙[1] 朱莉蓉[1] 

机构地区：[1]广西大学计算机与电子信息学院,南宁530004

出　　处：《计算机科学》2014年第9期104-109,共6页Computer Science

基　　金：国家自然科学基金(61063012;61363003);广西自然科学基金项目(2012GXNSFAA053222);广西高校优秀人才资助计划([2011]40);广西科学研究与技术开发计划项目(桂科攻1348020-7;桂科软13180015)资助

摘　　要：传统的串行模糊聚类分析算法在应对高维矩阵运算时存在运算量大、运算效率低等问题,难以满足云环境中集群资源调度的时效性要求。为此,在基于等价关系的模糊聚类算法基础上对传递闭包法进行优化,提出一种基于多线程的云资源模糊聚类划分并发算法,并将其应用于Hadoop调度器的策略改进。仿真实验结果表明,优化策略有助于减少平方法求解模糊等价矩阵的计算量,所设计的并发算法能够有效解决中小规模云集群资源聚类的运算瓶颈问题,且具有较好的加速比。为了解决现有Hadoop调度器存在的异构性问题,对该优化并发算法进行了理论分析,结果表明它有助于解决异构性带来的调度难题。The classic fuzzy clustering serial algorithm has the problems of heavy computation and low efficiency in dealing with high dimensional matrix operations, so it can＇ t be applied effectively into the fuzzy clustering partition model in cloud computing environment, and it＇s hard to meet the time efficiency requirement of resource scheduling. Therefore, a transitive closure method was optimized based on the equivalence relation-based fuzzy clustering algorithm. What＇s more,a fuzzy clustering concurrent algorithm based on multi-threading for cloud resources was applied to the improvement strategies for Hadoop scheduler. The experimental results indicate that the optimization strategy can reduce the computation for solving square-based fuzzy equivalent matrix problem. Moreover, the concurrent algorithm can effectively solve the computation bottleneck of resource clustering on small and medium-sized clusters,and it has a better speed-up ratio. To solve the problem of heterogeneity which exists in the existing Hadoop schedulers, theoretical analyses of the concurrent optimization algorithm show that it can help to solve scheduling problems caused by heterogeneity.

关 键 词：模糊聚类 云计算 资源聚类 模糊等价矩阵 HADOOP 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]