【導(dǎo)讀】師的指導(dǎo)下進(jìn)行的研究工作及取得的成果。盡我所知，除文中特別加。而使用過(guò)的材料。對(duì)本研究提供過(guò)幫助和做出過(guò)貢獻(xiàn)的個(gè)人或集體，均已在文中作了明確的說(shuō)明并表示了謝意。究所取得的研究成果。除了文中特別加以標(biāo)注引用的內(nèi)容外，本論。文不包含任何其他個(gè)人或集體已經(jīng)發(fā)表或撰寫的成果作品。本人完全意識(shí)到本聲明的法律后果由本人承擔(dān)。版，允許論文被查閱和借閱。本人授權(quán)大學(xué)可以將本。印、縮印或掃描等復(fù)制手段保存和匯編本學(xué)位論文。涉密論文按學(xué)校規(guī)定處理。

　　

【正文】 lity that the given state will lead to successful execution in the context where each Web service may fail with some probability. We show that the putation of aggregated reliabilities is equivalent to eigenvector putation and adopt the power method to efficiently derive aggregated reliabilities. In orchestrating a posite Web service, we propose two strategies to select Web services that are likely to successfully plete the execution of a given sequence of operations. A prototype that implements the proposed approach using BPEL for specifying the invocation order of a Web service is developed and serves as a testbed for paring our proposed strategies and other baseline Web service selection strategies. Index Terms—Reliability, service position, Web services, Web service selection. 1 INTRODUCTION THE Inter today has radically changed how businesses interact with applications distributed within and across anizational boundaries. With Web service (WS) technologies fast being the de facto standard to expose the functions of business applications, WSs will be the building blocks for developing the next generation applications using he serviceoriented architecture (SOA). Although the current WS architecture supports registry, discovery, and consumption of WSs, how to effectively integrate several WSs into a posite one remains to be a challenge and attracts much attention from both industry and academia. The position of WSs involves using an orchestration model to 1) define the possible orders of calling WSs at design time and 2) dynamically select WSs to be invoked at runtime. To address the former, several theoretical orchestration models have been proposed in the literature, including finite state machine (FSM), Petri , calculus, activity hierarchies, and rulebased orchestration [2]. Practical service position languages, such as BPEL, WSChoreography, WSCL, XPDL, and OWLS have also been proposed, and many mercial service platforms or products that implement the available standards and position languages are available in the market, including the Sun ONE framework based on J2EE, , the Oracle BPEL Process Manager, the HP WSs Management Platform, and the IBM Web Sphere Application Server. However, none of the 25 above works provide strategies or mechanisms to dynamically select WSs to be invoked when executing a series of operations in a posite WS. While deciding on a set of constituent WSs for a posite WS a priori is simple and practically used by most of today’s applications, it promises the flexibility and is especially vulnerable to failures in a dynamic environment on which WSs are executed. Dynamic WS selection refers to choosing the available WSs to be invoked so as to realize the functionality of a posite WS constructed using an orchestration model at runtime. While the WS selection problem has been extensively studied in the past few years, and various QoS measures, such as response time, cost, reliability, avail ability, and fidelity, have been proposed as the criteria for selection [15], [19], [20], [30], most works regard a WS as the basic unit for position. In reality, however, a WS may prise multiple operations such that their invocation sequence is constrained. A typical constraint is separation of duties, which is a wellestablished practice in security applications. In our context, separation of duties dictates that two operations must be delegated to two different WSs to avoid potential fraud [5]. On the other hand, an opposite constraint binding of duties may be needed under other circumstances. For example, we often mandate orders and payments to be made through the same WS. In addition, an enterprise may impose further constraints on the execution orders of its published operations to conform to its business logic. It is imperative that the dynamic WS selection procedure takes all these constraints into account. Researchers in previous works on automatic WS position [3], [7] therefore consider each service operation as the basic unit to pose aWS, and an FSMhas been adopted to specify the order among operations in a given WS. In addition, when choosing the operations of WSs to pose a (posite) WS, the atomicity of each WS, which requires either none of its operations to be invoked or some final state to be reached, has to hold at the end ofWS position. The goal of automatic WS position is therefore to determine whether a given posite WS can be derived using a set of WSs, and if so, the steps to construct the posite WS .While these works study automatic WS position as a decision problem, they have not addressed research issues arising from dynamically selecting candidate WSs to derive a posite WS so as to optimize some QoS criteria. The work (intra, extra, or even the Inter) where WSs operate is a failureprone environment partly due to the autonomous requirement of each participating service provider. A WS may bee malfunctioned or unavailable at runtime, causing failure to the execution of a posite WS. The analysis and putation of (software) reliability has been extensively studied in the past, and WS reliability can be measured from various perspectives, including correctness, fault tolerance, testability, interoperability, and timeliness [32]. In this paper, we define the reliability of a WS operation as the probability that it successfully responds within a reasonable period of time. We address the reliability issue of dynamic WS selection by extending the model adopted by Berardi et al. [3]. Note that WS is a stack of technologies, and some technologies are still under standardization process and are subject to change. Hence, the proposed approach is not restricted to any WS modeling language and employs a generic formalism FSM to define the order of operations in a WS. The order of operations in a WS can be intrinsically