分布式数据流处理系统的动态负载平衡技术  

Dynamic Load Balancing Techniques for the Distributed Stream Processing Systems

作　　者：邓华锋[1] 刘云生[1] 肖迎元[2] 

机构地区：[1]华中科技大学计算机学院,武汉430074 [2]天津理工大学计算机科学与工程系,天津300191

出　　处：《计算机科学》2007年第7期120-123,共4页Computer Science

基　　金：国家自然科学基金项目(60073045)资助;国家"十五"国防预研基金项目(413150403)资助

摘　　要：设计了一种新的大规模分布式数据流处理系统的体系结构。系统由一组异构的服务器集群组成,负载在每个服务器集群内部多台同构的服务器之间获得平衡,从而达到整个系统的负载平衡。集群设计的主要目标之一是以资源换性能,服务器集群中服务器的最大数目足够保证系统不再发生过载现象,不再需要会降低性能的卸载技术。而且投入运行的服务器的数目根据实际的系统负载来决定,负载较轻时,一部分服务器可以进入休眠状态来减少能源的消耗。根据系统动态增减服务器的特点,设计了全新的初始化算法、动态负载平衡算法。与以前的分布式数据流处理系统相比,由于单个集群的服务器的数目大大减少,算法复杂性降低、速度加快、优化的空间增大。In the novel architecture for the large-scale distributed stream processing system, the whole system consists of a group of heterogeneous computer clusters. The whole system can achieve the global load balancing by balancing every cluster which consists of several homogeneous servers. The main goal of every cluster is exchanging the resources for the performance. In the cluster, enough servers are employed to get rid of the occurrence of overload phenomenon, so techniques for load shedding are not still necessary in the system. In the meanwhile, the number of active servers is decided by the practical load level and some servers can be put into the sleep mode for the sake of energy conservation when the load is rather low. The ban＆new initialization algorithm and dynamic load balancing algorithm are designed to accommodate the characteristic of increasing or decreasing the servers dynamically. Comparing to the traditional largescale stream systems, these algorithms have better load balancing, lower complexity and faster response time because the number of servers in a single cluster is reduced sharply.

关 键 词：分布式数据处理流系统 动态负载平衡 卸载 节能 

分 类 号：TP311.138[自动化与计算机技术—计算机软件与理论]