【作者】 李晨；

【导师】 孙涌；

【作者基本信息】 苏州大学 ， 计算机应用技术， 2007， 硕士

【摘要】 集群的动态负载均衡是集群计算中的关键技术，如何提高动态负载均衡的性能，一直是集群计算研究的热点。本文研究的集群负载均衡问题，其目的就是在互联网用户数和网络流量呈几何级数增长的情况下，使集群系统仍然能够提供高可用高扩展性的服务。本文首先分析了集群技术及其负载均衡技术的现状和发展趋势，指出了当前集群负载均衡机制存在的问题和不足之处，在综合分析了集群计算技术和Agent技术的基础上，将Agent技术引入集群负载均衡机制中。提出了将Agent的强化学习机制和负载均衡转移技术相结合的负载分配策略，并设计了一个基于多Agent系统的集群负载均衡模型。本文对该模型的设计思路、体系结构和工作原理都进行了较详细的论述。文章结合Web层容器集群的特点，在其基础上实现了该模型。并验证了其性能效果优于传统的集群负载均衡算法。文章最后对基于多Agent系统的集群负载均衡研究和设计做了总结，并提出今后需要进一步研究的内容。本文在Agent在集群技术应用方面做了一些工作，如强化学习在负载均衡策略中的应用，多个Agent如何相互协作共同完成系统功能等。这些内容对集群系统中相关研究有一定的参考价值，具有一定的借鉴意义。更多还原

【Abstract】 As the dynamic load balancing is the crucial technology in clustercomputing, how to improve the performance of dynamic load balancingis one of the focus areas to its researchers. For the Internet users and thenet flow follow the geometric series increase, it requires that clustersystem should have high availability. It is the purpose of this paper to findsome way to solve these problems.Firstly, this paper discusses the present situation and development ofcluster and load balancing technology, and brings forward the problemsand limitation of current cluster load balancing mechanism. Then, basedon the synthetically analysis of agent technology and cluster computingtechnology, we integrate the intelligent agent with cluster load balancingmechanism. We present strategies which integrate agent reinforcelearning and load transfer technology, and design a cluster load balancingmodel based on multi-agents system. The model’s targets, architectureand working principle are discussed subsequently in detailed. After that,we implement this model on the base of Web Container Cluster.Compared to traditional static algorithms of cluster load balancing, ourmodel produced higher performance in the test cases. At last, this paperdraws a conclusion of cluster load balancing based on multi-agentssystem research and design, and putting the further work in the future.The main contribution of the paper is to apply agent technology tocluster load balancing mechanism, such as reinforce learning in the loadbalancing strategies, how multi-agents work together to achieve systemtarget, etc. It is worthwhile that these technologies are helpful to thedevelopers in cluster field.更多还原