【作者】 沈华；

【导师】 何炎祥；

【作者基本信息】 武汉大学 ， 计算机软件与理论， 2013， 博士

【摘要】 Web服务组合的本质是通过若干Web服务间的相互通信和协作灵活实现上层业务需求,并能够屏蔽底层实现技术的改变对上层产生的影响。Web服务组合是面向服务架构(Service-Oriented Architecture, SOA)中构建新服务和满足业务需求的主要手段,在业务流程的构造和重组中处于核心地位,因此Web服务组合的性能及其评价方法对SOA的发展至关重要。本文结合国家863项目“动态Web服务组合关键技术及其性能分析研究”,从定性分析和定量分析两个角度对Web服务组合的性能评价展开讨论,对其中的关键技术进行了研究。本论文的主要工作包括以下内容：(1)为了在Web服务组合从设计产生到服务结束整个生存期的各个环节监控其质量,提出一种贯穿Web服务组合的整个生存期的Web服务组合性能评价体系结构。该体系由两个闭环负反馈系统组成,分别对Web服务组合做了定性性能分析和定量性能分析。为了实现该性能评价体系,提出一种基于随机Petri网的Web服务组合的性能分析模型WSCPAM,并分析出它天生具有的两个结构约束。WSCPAM是性能评价体系的物质基础,体系所开展的评价工作均在其上进行。(2)为了能够在设计阶段发现结构中固有的致命弱点,保证运行时的Web服务组合是良结构的,给出WSCPAM进行各项结构验证(包括结构约束的验证、活性的验证、有界性的验证、死锁的验证和陷阱的验证)的意义,并给出实现算法。(3)为了有效利用WSCPAM模型对Web服务组合进行性能的定量分析,给出了一个详细完整的WSCPAM模型转换为同构马尔可夫链的算法。算法主要包括三个规则：转换规则、演变规则和合并规则。运用Floyd不变式断言法证明了算法的部分正确性,运用良序集法证明了算法的终止性。算法是部分正确的且是可终止的,这个证明结论说明算法是完全正确的。对算法时间复杂性的分析表明算法是有效的。利用产生的马尔可夫链对WSCPAM模型的多项性能指标进行定量分析。(4)Web服务组合的性能对推动以服务为核心的价值链的良性融合起着重要作用。针对如何有效发现Web服务组合中性能瓶颈的问题,提出两种Web服务组合性能瓶颈定位策略BNL-WSCPAM和BRF-WSCPAM。根据库所和变迁之间的直接关系总结出模型的四种基本结构,并证明了它们的完备性(即四种直接关系之间的不同复合可以表示出Web服务的各种组合方式)。通过结构转换得到性能瓶颈定位分析结构集{顺序结构、分支结构、汇聚结构},在此基础上提出性能瓶颈定位策略BNL-WSCPAM。该策略根据与时间有关的性能指标对库所进行分类,针对不同类型库所在三种分析结构中的所有可能分布,通过顺向结构分析和逆向结构分析相结合的方法,讨论瓶颈可能存在的位置,并给出如何确定三种分析结构的结构发现定理。实验证明策略的预测结果是有效的。与BNL-WSCPAM不同的是,性能瓶颈定位策略BRF-WSCPAM是仅仅基于性能指标提出的。在前面求得的各项性能指标中分析出对产生瓶颈贡献大的性能指标,将这些性能指标做归一化的处理,得到一个综合的性能参数——瓶颈风险系数,针对这个参数对模型中所有的变迁做一个统一的度量,从中发现可能存在的瓶颈。更多还原

【Abstract】 The nature of Web services composition is to flexibly achieve upper business needs through communication and collaboration between Web services, and can mask the impacts which are the underlying implementation technology changes haven on the upper. Web services composition is the primary means to build new services and meet business needs in Service-Oriented Architecture (SOA), and is at the core position in business processes’ construction and reorganization. Therefore, the performance and performance evaluation methods of Web services composition are important to the development of SOA.Supported by the National High-Tech Reaearch and Development Plan (863) of "the researches about key technologies and performance evaluation of dynamic Web services composition", we discuss on the performance evaluation of Web services composition from two views which are qualitative analysis and quantitative analysis, and we research on the related key technologies. The main contributions are as follows:(1) In order to monitor the quality of Web services composition in all aspects of the period from Web services composition’s generation to its end, this paper proposes a kind of performance evaluation architecture for Web services composition which is throughout the entire lifetime of Web services composition. The architecture consists of two closed-loop negative feedback systems. The one provides qualitative analysis for the performance of Web services composition, and the other one offers quantitative analysis for the performance of Web services composition. In order to achieve the performance evaluation architecture, this paper proposes a performance analysis model WSCPAM for Web services composition, which is based Stochastic Petri Net. And the two inherent structure constraint conditions of WSCPAM are discovered. WSCPAM is the material basis of the architecture, on which all performance evaluation jobs by the architecture are carried.(2) To be able to discover the inherent Fatal weaknesses of the structure in the design phase and ensure run-time Web service composition is good structure, this paper gives the meanings of conducting various structural verifications (including structural constraint conditions verification, liveness verification, boundedness verification, deadlock verification and trap verification), and provides algorithms to implement these verifications. (3) In order to effectively using WSCPAM model to conduct performance quantitative analysis for Web service composition, this paper presents a detailed and complete algorithm for converting WSCPAM model to isomorphic Markov chain. The algorithm mainly includes three rules, which respectively are conversion rule, evolution rules and merger rules. The Partial correctness of the algorithm is proved by Floyd invariant assertion method. The Termination of the algorithm is proved by well-ordered set method. The algorithm is completely correct, which can be illustrated by the conclusion which is that the algorithm is partial correct and can be terminated. The time complexity analysis of the algorithm shows that the algorithm is effective. Using the generated Markov chain, a number of performance indicators of WSCPAM model can be calculated.(4) The performance of Web services composition plays an important role in improving the virtuous integration of the value chain at the core of services. For how to effectively identify the performance bottlenecks of Web services composition, this paper proposes two performance bottlenecks location strategies BNL-WSCPAM and BRF-WSCPAM.Based on the all direct relationships between place node and transition node, four basic structures of the model are summarized. And the completeness property of the four basic structures is proved, which means the different compositions between the four kinds of direct relations can represent various Web services combinations. Through structural conversion, an analysis structures set is obtained, which includes Sequence structure, Split structure and Join structure, and is used to predict performance bottlenecks that may exist. On the basis of the set, the paper presents the performance bottlenecks location strategy BNL-WSCPAM for Web services composition. According to time-related performance indicators, the strategy classifies different types of places. Under the various distributions of different types of places in three kinds of analysis structures, the performance bottlenecks that may exist have been discussed through the Down Arc Analysis method and Reverse Arc Analysis method. This paper also gives Structure Discovery Theorem, which is used to determine the three kinds of analysis structures. The experimental results show that the prediction of the strategy is valid.It is different from BNL-WSCPAM that the performance bottlenecks location strategy BRF-WSCPAM is solely based on performance indicators. Through analyzing the obtained performance indicators in previous works, it finds these performance indicators which make large contribution for bottleneck generation. And then through normalize these indicators, a comprehensive performance parameter-Bottleneck Risk Factor is obtained. Based on the parameter, all transitions in WSCPAM are uniformly measured, and some bottlenecks may be found.更多还原