【作者】 王开宇；

【导师】 田乃硕；

【作者基本信息】 燕山大学 ， 电路与系统， 2010， 博士

【摘要】 随着互联网的发展,Web-World Wide Web(WWW)的应用领域更加的广泛,技术发展更是多样化。对于迅速增长的网络用户,Web必须能够提供更加可靠和高质量的服务,才能应对当前的发展,所以对于Web系统的性能分析是一个重要的研究课题。在深入分析并掌握国内外研究现状的基础上,采用排队论和排队网络数学理论,对大型的Web集群系统建立模型,对常见的多层Web应用基础平台建立模型,对Web应用程序建立模型,得到系统的性能指标表达式。通过仿真和测试的手段进行比较分析,给出系统优化设计,对Web系统的性能规划和应用软件系统的设计提供了理论依据和分析方法。首先,采用基础的M/G/1排队模型,给出集群系统在考虑服务器性能和数量时的优化搭建方案;通过把复杂排队网络进行转换,基于M/M/c排队模型进行分析,给出集群系统中优化的服务器数量范围;采用异步休假的M/M/c排队模型,给出集群系统优化的能耗管理方案;采用仿真的方法对集群系统的负载均衡策略进行了性能分析。对相关模型进行了数值分析与验证。其次,基于多层的Web应用程序运行环境,分别对Web服务器、应用服务器和数据库服务器建立模型,通过排队网络的模型分析,与系统测试相结合,给出服务平台构建的优化配置方案。对于B/S模式下的Web基础网络平台建立了排队网络模型,采用排队网络模型MVA求解方法和扩散近似的方法给出了性能解析表达式,通过NS-2进行系统仿真对比,分析系统瓶颈并给出改进方案。再次,对集群环境下企业级的J2EE应用程序进行了排队网络模型的分析,对小型的共享主机平台的J2EE应用程序进行排队网络模型分析,通过测试验证了模型的正确性,给出系统的优化搭建方案。对于组合Web服务,按照并行组合关系建立马尔科夫链模型,对复杂的组合关系建立排队网络模型,分别进行求解并进行数值分析,给出系统优化的构建方案。对Web应用软件的UML设计方案进行模型的转换,采用分层排队网络模型进行分析,通过分层模型的MVA算法,得到系统相关的性能指标,对软件性能进行预测分析。最后,通过对实例Web应用中的AJAX技术进行响应时间和网络流量的分析,给出Web应用软件的实际性能优化方案,对其他影响性能的相关因素也进行了简要分析。更多还原

【Abstract】 With the development of Internet, Web- World Wide Web(WWW) ’s applicable field become more extensive, the progress of technology tends to be diversified. For the quickly increasable web users, Web must provide more reliable and high quality services, so it can keep up with the development. It is a very important research field for the web system performance evaluation. Based on deeply analyzing and mastering current research state at home and abroad, using the mathematic theory of queueing and queueing network, modeling the large web cluster system, modeling the basic platform of the multi tier web application, modeling the web application. The system’s expressions of performance measures are obtained from simulation and tests. Through comparing and analyzing, the system’s optimal results are presented, which provides theoretical basis and analysis methods for the web system performance planning and Web application designing.Firstly, by adopting the queue model of M/G/1, the optimal establishing policy of the cluster system are presented, considering the servers performance and server number simutaniously; Through the transition of complex queueing network model, by analyzing based on the M/M/c queue model, the range of optimal server number of the cluster are obtained; By adopting the M/M/c queue model with vacations, the optimal power management policy of the cluster are presented; from simulation, performance evaluation is done for the cluster’s load balancing policies. Numerical analysis and validation is done for these models.Sencondly, based on the running enviroment of the multi tier web application, models are presented for web server, application server and database server respectively. Through queueing network analyzing, integrating the system tests, the optimal configuration policy of the sevice platform’s establishment are presented. For web’s basic network platform, queueing network model is given. By adopting the MVA algorithm and diffussion approxmimate algorithm of queueing network, the performance expressions are obtained. By comparision with simulation of NS-2, system’s bottleneck is analyzed and improved policies are presented. Thirdly, queueing network analysis is done to the large scale J2EE application established in server clusters and small J2EE application on sharing hosts. Model’s accuracy is validated from tests. The system optimal establishment policy is presented. For the composite web services, a marchov chain model is established according to the parelell relationship. A queueing network model is established according the complex composite relationship, Solving and numerical analysis are done respectively. System optimal establishment policy is presented. Through transition for web application’s UML model, by analyzing from layered queueing network, according to the MVA algorithm of layered queueing network, some performace indices of the application are obtained, so performance prediction are done to the software.Finally, through analyzing of response time and traffic flow of AJAX of the instances of web application, actual optimal performance policies are presented for the web application, simple analysis for other performance indices are also presented.更多还原