【正文】 ransaction/exception handling. Meanwhile, several service position languages and process specification languages have been developed. Some of them originate from the domain of WS (., BPEL, WSCL, and WSChoreography), some from the Semantic Web (., DAMLS/OWLS), and others from the development of a standard format (., XPDL) for data exchanges between different Workflow Management Systems (WFMSs) with inpatible internal process models. In [27], van der Aalst et al. reviewed various workflow patterns that cover most of the scenarios ever described in realworld workflows and concluded that not a single process specification language could specify all the possible interaction patterns.Several formal process metamodels have also been proposed in the literature, mainly for analysis purposes, including FSM [3], Petri net [22], activity diagram [7], and process algebra [8]. In [20], Milanovic and Malek pared several popular WS position languages and formal models and concluded that an FSMis a promising model for analyzing service connectivity, position correctness, automatic position, and scalability. Moreover, an FSM is a formalism that is suitable to describe reactive behaviors and has the notion of states, which is useful for monitoring service executions. In addition, an FSM can be used to describe the underlying logic of Petri net, another popular formal process metamodel. As a result, an FSM bees the most popular formal model for describing operation orders in a WS and has been adopted by some position languages (., WSCL and WSChoreography). Severalapproaches for converting a BPELWS to an FSM representation have been proposed [9], [11]. Web Service SelectionThe WS selection problem has been extensively studied in the past few years. Previous works have focused on optimizing the selection of WSs for a single activity, while the most recent ones focus on the selection of WSs in order to satisfy the QoS requirements of a workflow (or posite（WS). To optimize the selection for the entire process, Zeng et al. [29] proposed puting an optimal set of WSs for each possible execution path in the process based on a weighted bination of QoS measures. When it es to deciding a WS for a given activity t, the most popular execution path on which t appears is chosen, and the WS assigned to t in that path is selected. To speed up the putation, techniques of integer programming are employed. However, the number of possible execution paths could be huge, especially in the presence of a loop. Besides, the failure of WSs at runtime requires the repilation of a set of new integer programs, which may not be feasible in practice. Yu et al. [28] proposed a scheme to optimize the endtoend QoS for various flow structures. They used a utility function derived from the QoS quality and formulated the optimization problem as a general flow problem. Some works concentrate only on a single QoS measure. Menasce [19] proposed a scheme to estimate the throughput of a posite WS from those of its constituent WSs and to use throughput as a basis for selecting WSs. Grassi and Patella [13] proposed a frame work to recursively aggregate the reliability of a posite WS based on those of its constituent WSs. Several works have been devoted to the derivation of other QoS measures, such as response time, cost, availability, and fidelity, of a posite WS out of those of its constituent WSs [6], [15].All of the abovementioned works adopt a rather simple model that views a WS as the basic unit for posing a posite WS. Our work adopts a different model in which every WS (posite or ponent) prises a number of operations, and the invocations of their operations may have to follow some particular orders. In addition, our proposed approach does not need to enumerate all possible execution paths before a process is initiated and allows partial reputation when a candidate WS is found to fail at runtime, potentially incurring less putation time. Web Service CompositionOne notable research topic that has attracted much attention recently is automatic WS position [3], which is aimed at verifying the feasibility of posing a posite WS using a munity of candidate WSs. Each (posite or candidate) WS is associated with a formal model that constrains the order of its operations. Notable models include Petri net [18], [22] and FSM [3], [11], [12]. While each operation in the posite WS can be realized by some candidate WSs, additional constraints, as imposed by the respective formal models, have to be satisfied. Typical constraints include the atomicity (all or nothing) of each WS [3], [11], the patibility of the constituent WSs [18], QoS requirements [1], and that all operation sequences of the posite WS must be captured [3]. The objective of automatic WS position is to determine a position that satisfies all the given constraints. Web Service ReliabilityThe reliability of a WS refers to the probability that the WS will successfully execute. Several works have been proposed in the literature to derive the reliability of aWS in a dynamic environment. In [32], a methodology was proposed to effectively test the reliability of aWS by generating test cases of different categories. As the reliability of aWS may change over time, an architecturebased reliability model has been proposed [26] to incrementally update the reliability of an atomic WS by continuously generating test cases and aggregating the test results using a voting scheme. The reliability of a posite WS can be derived by aggregating the reliabilities of constituent WSs based on the occurrence rate of each flow pattern. While a thorough test on a (remote) WS may require intensive interaction between the WS and the test requester, Zhang proposed a mobile agent based approach that uses a mobile agent to perform the onsite test for WS so as to reduce the network traffic [31].3 PROBLEM DESCRIPTIONIn this section, we articulate the key concepts for technical discussions, whi