【作者】 易琦娜；

【导师】 晏蒲柳；

【作者基本信息】 武汉大学 ， 通信与信息系统， 2004， 硕士

【摘要】 目前，网络应用的增长速度高于目前网络技术本身的发展速度，不可避免的存在处理延迟和拒绝服务现象，这就使得端到端的服务质量保证显得越来越重要。端到端QoS包括网络QoS和端系统QoS两个方面。电子商务应用的日益发展和HTTP请求的指数性增长造成Web服务器中的拥塞和排队现象越来越严重，仅靠网络QoS机制不可能完全解决端到端的QoS控制问题。Web服务器作为端到端网络中不可缺少的一个重要环节，必须具备支持QoS的机制和策略。 如何在Web服务器系统中引入和实现QoS控制的机制与策略，满足不断增长的性能需求，为不同类型的用户或请求提供服务区分和性能保证，是当前Web研究发展的一个热点问题。 网络的迅速发展造成人们对Web服务器的高可伸缩性、高可用性、易管理性、高性能价格比的要求越来越高。Web服务器集群系统成为目前满足上述要求的一个既经济又有效的手段。从而Web服务器集群的服务质量保证是为用户提供满意的Web QoS的一个重要方面。Web服务器集群的QoS研究正处于起步之中，关于其的研究论文国内外都不多见，更别说产品和专利技术了，本文的研究即以此为出发点和主要目标。 本文首先研究了QoS和Web QoS控制的一些背景知识和当前的研究状况，得出Web集群服务器的QoS研究的必要性。然后在研究了服务器集群的QoS模型基础上，总结出两种集群QoS模型的方式——分布式QoS策略服务器集群模型和混合式QoS策略服务器集群模型。再用排队论理论对两种模型进行了比较，得出分布式QoS策略服务器集群模型要优于混合式QoS策略服务器集群模型。 随后本文提出并实现了基于分布式异构实时的QoS策略-DAHRP的WebCluster QoS模型，该模型包括性能监测部件、接纳控制部件、请求分级部件、服务器分配部件、Web页面调整部件和请求调度部件。文中对实现中的关键问题进行了详细的描述。最后在设计好的Web Cluster QoS模型上，根据自行提出的模拟现实Web访问特点的服务器性能测试仿真算法，进行了防真试验。比较了本文提出的DAHRP策略与现有的QoS策略的差别，证明了其优越性，也证明了我们所作的工作无论在理论上还是在实际应用中都是很有意义的。更多还原

【Abstract】 The growth of the demand of the network is higher than the development of network itself at present. So it is still unavoidable to deal with delay and denial of service phenomena. This makes end-to-end Quality of Service (QoS) seem more and more important. And end-to-end QoS includes two respects: networks QoS and end-system QoS. With the growth of E-commerce and the increase of HTTP requests, the phenomena of congestion and queueing are more and more serious. So it is imposible to solve the problem of end-to-end QoS control by only depending on networks QoS. And thus Web Server becomes an important part in end-to-end networks. It is necessary to study the QoS mechanisms and policies for it.With the growth of network, high scalability, reliability, availability and efficient cost/performance ratio are strongly required for the network servers. Web server cluster is an economic and effective method that can meet the demand. So Web server cluster QoS is an important respect of offering satisfying Web QoS to users. The papers about it are not many, and let alone the products and patents. This paper regards this as the starting point and main goal.First this paper studies background knowledge and present research state of Web QoS control and the necessarily of studying the Web server cluster QoS. This paper makes the decision that there are mainly two server cluster QoS models: the server cluster QoS model of distributing QoS policy and the server cluster QoS model of mixing QoS policy. And after the comparison of the two models, this paper draws the conclusion that model of distributing QoS policy excels model of mixing QoS policy.And this paper proposes a Web Cluster QoS model in which a distributing-aware heterogeneous realtime policy -DAHRP is realized. This model includes performance monitoring part, admission control part, requests classification part, servers partition part, web adjustment part and requests scheduling part. The key problems have been detailed described in this paper. In the end, this paper makes a simulation on the Web Cluster QoS model using a simulation algorithm that is proposed by oneself. From the comparison’s result, it is proved* that DAHRP is better than the others.更多还原