【作者】 冯兴智；

【导师】 吴泉源；

【作者基本信息】 国防科学技术大学 ， 计算机科学与技术， 2007， 博士

【摘要】 在开放、动态多变的Internet环境下,服务组合能够实现动态资源聚合和灵活应用集成,已成为面向服务计算(SOC)技术发展的主要方向之一。缺少服务质量保证,虽然通过服务组合能够完成用户的基本业务要求,但是往往不能达到用户的满意程度。特别是,由于业务目标的动态性和服务本身的自主性,如何在已保证功能性需求的前提下,应对实际应用的非功能性方面需求,并满足业务执行过程中的动态适应性要求,是基于服务质量的动态服务组合重点解决的问题。目前基于服务质量的动态Web服务组合在组件服务选择、组件服务组合相容性、动态执行中断恢复等方面尚面临诸多新的应用挑战。尤其对于要求高可用性和可靠性的核心任务和重要实时系统,最小中断的服务能力保证对用户来说至关重要。为此,本文的研究工作围绕基于服务质量的动态Web服务组合的服务选取、相容性检查、失效恢复等关键技术展开探索和研究。主要工作和贡献包括:(1)提出一种基于多维服务质量的服务选择模型。服务选择是服务组合过程中首要和必要的工作。服务的非功能属性尤其是服务质量成为面向服务计算范型中关注的重要方面,选择最佳的满足QoS约束的组件服务来构建组合服务至关重要。但是目前研究存在全面评价标准模糊、QoS参数赋权模式未考虑主客观因素等问题。为此,本文提出一种基于服务质量的服务选择模型,为服务选择以及动态执行中的服务替换提供多维QoS支持;定义服务质量度量子模型,给出一套具体的服务质量效用评价函数(服务效用函数和边效用函数);定义服务质量属性的权重子模型,使用决策模式基于用户主观感受和客观事实采用主客观赋权模式来设置QoS指标权重;提出一种基于MMKP背包问题的使用规则库的启发式选择RWSCS KP算法。实验表明主客观赋权模式公正合理,启发式选择算法通过使用规则库降低了搜索空间大小,能够快速有效的完成服务选择。该部分研究为动态服务组合的相关关键技术研究提供一致的服务质量计算框架。(2)提出一种基于判定规则的服务相容性模型。在服务选择阶段提供执行规划的合理性保证对于保证组合服务的正确执行至关重要,有助于减少运行期失效并降低组合代价。但是目前服务相容性模型研究还不全面充分,缺乏可操作性。本文提出一种服务相容性模型,从语法、语义和非功能属性三方面给出一套层次式相容性规则;定义了相容度的概念来表示组件服务间具体的匹配程度,通过相容度阀值来界定服务是否相容,并给出了具体的相容度松弛计算函数。最后实验考察了相容度阀值取值对组合服务成功率的影响并证明了在服务选择阶段进行相容性检查的必要性。该部分研究工作通过判定规则和计算函数来度量服务的相容程度,与相关工作相比具有较强的表达能力和可计算性。(3)提出基于执行环境上下文的服务路径动态替换算法。组合服务在实际执行过程中,由于组件服务和执行环境的不可控性容易发生执行路径中断,需要动态调整原有路径。本文提出一种上下文相关的动态路径替换WSCCDR算法,根据实际执行过程的上下文信息分别采用全局和局部路径替换方式,并针对实际业务流程执行已完成相当大部分的情形,重点研究了局部服务路径动态替换WSCPR算法,利用已完成的组件服务来缩短整个组合服务的执行时间。对于业务流程执行路径的多段关键子路径,提出K最大效用路径替换WSC KUPB算法,一次为多个关键路径阶段分别寻找一条替换路径。实验表明算法在不同测试集下近似线性的运行时间增长,能够快速有效地完成路径替换。并且通过还原成功率实验表明算法在一个适度变化的overlay网络中可成功进行大多数的失效还原。算法解除了目前相关服务路径动态替换算法对服务路径拓扑的限制,为类似的工作提供了一个新的思路。(4)基于服务质量的动态服务组合原型的设计与实现。基于上述关键技术的研究探索,在国防科技大学研制的基于Web Services分布应用支撑平台StarWebService基础上,设计实现了基于QoS的动态服务组合原型StarWSCX,对于动态服务组合支撑平台的体系结构和具体功能组件进行详细讨论,并重点关注了Web服务组合中基于多维QoS的服务选择和替换问题、相容性检查方法和策略,以及执行过程中的SLA问题和失效还原机制。更多还原

【Abstract】 In the open and dynamic Internet environment, service composition can realize dynamic resource aggregation and flexible application integration, and has been the major development trend of Service-Oriented Computing. But for the shortage of QoS provision, service composition can’t meet the user’s satisfaction, though it can meet the base business requirements. Especially for the dynamic nature of business objects and autonomy of service itself, the key problem is how to deal with the non-functional requirements in practice on the premise of provision of functional requirements and meet the demand of the dynamic adaptability during the business execution process.Current research in QoS-based dynamic service composition still faces many new application challenges in some important respects such as the selection of component service, service composition compatibility and interruption recovery during the dynamic executution, etc. The service capability provision of minimal disruption to the consumers is very important feature especially for high-availability and reliability, critical mission and hard real-time systems. Thereby in this dissertation the research work is focused on key technologies such as QoS-based service selection during dynamic service composition, compatibility checking and failure recovery. The contributions of this dissertation are as follows:(1) A service selection model with multiple QoS constraints. The service selection is the first and critical step during service compositon. The non-functional properties of service especially the quality of service are the key aspects in the service-oriented computing paradigm. It is important for business process to construct a complex one with multi-dimensional QoS assurance. But there exist some problems, such as evaluation for QoS properties of service is not full-scale and the criteria is not clear, the weight for each QoS metric doesn’t consider both subjective sensations and objective facts. In this dissertation we propose a service selection model to provide multi-dimensional QoS support in service selection and dynamic replacement. We define an evaluation sub-model of QoS, and propose a set of utility functions (Service Candidate Utility Function and Edge Utility Function) as the evaluation standards, which aggregate all QoS metrics after normalizing their values. Then we define an evaluation sub-model of QoS weights and use Subjective-Objective Weight Mode (SOWM) to set the weight of each QoS metric by means of decision mode. Also we propose a heuristic algorithm on the MMKP using the rule-library. Finally the experiments show that the SOWM is fair and rational enough and the algorithm is helpful to finish the selection quickly with the benefit of rule-library reducing search space at a certain extent. The research in this part provides a uniform computing framework of QoS for the key technology research in dynamic service composition.(2) A compatibility model with judgement rules for dynamic web service composition. It is very important to the correct execution of composite service with the reasonableness provision of execution plan in service selection stage, and it’s helpful to avoid the runtime failure and reduce the execution cost. The current research on compatibility model is not sufficient and short of manipulation. In practical it is not always possible to find services that are fully compatible with each other. Composers fail to ensure a perfect match in the composition of Web Services. In this dissertation a compatibility model is proposed to ascertain that Web Services can be safely combined, hence avoiding unexpected failures at runtime. Compatibility is checked through a set of rules organized into three levels: syntactic, semantic and non-functional properties levels. The concept of compatibility degree is introduced to cater for partial and total compatibility with the help of compatibility threshold, and a concrete relax function is proposed to embody the compatibility degree of Web Services as numeric value. Finally a simulation is conducted to test the impact of compatibility threshold on the success ratio of service composition, and the result proves the necessity for compatibility checking in the selection stage. This model evaluates the degree of compatibility between services by the judgement rules and computation functions. It has a more strong expression ability and calculability compared with the past research.(3) Efficient dynamic replacement path algorithms based on the execution context-oriented environment. During the execution of Web Service Composition, it is familiar that the running process is interrupted for the uncontrollability of component services and execution environment. A failure recovery mechanism is needed to ensure that execution path can be replaced quickly and efficiently. In this dissertation, we present a context-sensitive adaptive algorithm (WSCCDR) that can use global and local replacement methods respectively, based on the actual execution context. And in the case that most of the business process has been executed, we propose the WSCPR algorithm to find replacement path locally by virtue of the old executed component services to shorten the total execution time. With respect to multi-stage critical sub-paths in the business execution path, another algorithm WSC_KUPB is described to enumerate the k maximum utility simple loopless paths in a digraph, which can find a replacement path for every critical path at one time. The test’s result shows the algorithms perform very well that run time increases with the approximate linear feature under different test cases, and high failure recovery success ratio can be achieved on a moderately changing overlay network by our algorithms. Our algorithms have no restriction on the topology of the overlay network that current related algorithms require. And our work suggests a new way to develop efficient dynamic replacement algorithms for QoS-aware service composition. (4) Design and implementation of QoS-aware service composition prototype. Based on the above studies of the key technologies and the Web Services supporting platform named StarWebService developed by National University of Defense Technology, we propose a prototype StarWSCX for Qos-aware dynamic service composition. The architecture of dynamic service composition supporting platform is presented and the concrete components are discussed in details. The key issues such as QoS-based service selection and replacement, the methods and stategies for compatibility checking, SLA and failure recovery mechanism during the execution process are emphasised.更多还原