【作者】 付晓东；

【导师】 邹平；

【作者基本信息】 昆明理工大学 ， 管理科学与工程， 2008， 博士

【摘要】 在开放的网络环境下实现跨组织的信息共享与业务协同已成为商业、科学研究、军事等各个领域中具有广泛需求的基础性研究课题。近年来,随着“服务”成为开放网络环境下资源封装与抽象的核心概念,通过动态地组合服务实现资源的灵活聚合成为技术发展的自然思路。特别是随着Web服务技术的出现和推广,Web服务已成为公认的实现服务的主流技术选择,这使得动态Web服务组合技术成为面向服务的计算的核心技术。由于应用领域的多样性、复杂性以及用户需求的动态性,Web服务组合需要具有服务动态发现、选择与绑定的能力。然而,运行环境的动态性和不可预知性使得Web服务组合服务质量(Quality of Service, QoS)保障成为至关重要且意义重大的挑战。目前,许多国内外研究机构围绕Web服务组合服务质量保障展开研究工作,并取得了一些有价值的探索性成果。但是作为一项新兴的研究课题,Web服务组合质量保障在服务信誉度量、服务选择、故障诊断以及异常处理等方面尚不成熟,新的研究问题不断涌现。本文以开放网络环境下的资源共享与应用集成需求为背景,展开Web服务组合质量保障研究。在对当前服务组合质量保障研究中存在的若干不足进行分析的基础上,重点在Web服务信誉度量、基于QoS的Web服务选择、Web服务组合故障诊断以及异常处理等几个方面展开研究工作,主要包括：1.信誉是判断QoS公告信息真实性的基础,从而也是Web服务选择的基础。针对目前信誉度量多依赖于用户主观评分的情况,结合Web服务自身的特点,提出基于质量实际值与公告值之间相似度进行Web服务信誉度量的模型。首先通过上、下两层本体描述了Web服务QoS的语义和层次化的QoS指标体系,然后设计一种支持层次化QoS指标体系的质量相似度算法,并将质量相似度作为信誉度量的依据。信誉度量算法以统计理论为基础,通过引入遗忘因子和波动因子,考虑了服务质量的发展趋势和波动情况对服务信誉的影响。实验表明该模型具有良好的灵活性和实用性。2.如何动态地选择出最适合消费者QoS需求的Web服务是Web服务在商业应用中获得成功的关键。目前的研究多采用精确的服务质量信息进行服务选择,没有充分考虑到服务选择中QoS信息的不确定性,在对Web服务选择中的信息不确定性进行分析的基础上,设计了QoS公告值、需求值及权重约束的不确定表示方法。然后,利用区间数比较的可能度方法和逼近理想点的多属性决策方法,建立基于不确定QoS信息的Web服务选择模型。实验表明,该模型在QoS信息不确定的情况下能有效地选择符合消费者需求的Web服务。最后,通过聚合树实现结构化Web服务组合QoS的聚合,并提出Web服务全局选择的算法。3.为有效识别引起组合服务故障的组件服务,提出一种利用贝叶斯网络对Web服务组合故障原因进行诊断的模型。首先对模型中贝叶斯网络拓扑结构的构造以及网络参数的赋值方法进行了详细说明。然后,基于构造的贝叶斯网络,设计了Web服务组合的故障诊断算法并对其进行了分析。最后,根据信息熵及条件熵的概念,提出基于k-median的证据通道选择算法,以降低收集观察证据的成本。实验模拟表明,提出的模型能有效地识别Web服务组合流程错误的根本原因。与已有的研究相比,该故障诊断模型考虑了故障诊断推理过程中的不确定性以及证据通道选择的问题。4.为实现Web服务组合异常处理的模块化、可扩展性及适应性,提出一种基于元流程的Web服务组合异常处理模型,并对该模型的元素及其关系语义进行了详细说明。模型中的元流程是一种高层控制流程,它由相应的规则启动,并对基础流程进行协调和控制。该方法在异常事件发生时,不是简单地引发一个活动,而是通过相应的规则驱动元流程对基础流程进行协调和控制,从而以一种较简单的方式实现对组合服务异常的处理。实例分析表明该方法不仅支持错误处理,同时也支持流程修改等复杂状况。5.在前面研究成果的基础上,提出一种保障服务质量的Web服务组合参考模型,模型体现了本文对于服务组合质量保障体系的理解,为Web服务组合支撑平台的建立以及基于Web服务组合的应用系统构造提供了一个可供借鉴的计算框架。综上所述,本文的工作针对目前Web服务组合质量保障中亟待解决的几个关键问题提出了有效的解决方案,对于推进动态服务组合技术的理论研究和实用化具有一定的理论价值和应用价值。更多还原

【Abstract】 Resource sharing and application integration across organizational boundaries in open network have been recognized as comprehensive requirements in many domains, such as commerce, scientific research, military area. In recent years, the concept of service has been proposed as a mean of abstracting and wrapping diverse resources in open environment, which makes service composition to be the natural way for dynamic resource aggregation and flexible application integration. Especially, along with the prosperity and acceptance of Web services as the mainstream technology for service provision, dynamic service composition based on Web services has been regarded as one of the core technologies for the service-oriented computing paradigm. Due to the diversity and complexity of applications, as well as the complexity of the user’s dynamic requirements, Web Services need to have the ability of dynamic service discovery, selection and binding. However, dynamic and unpredictable environment makes the QoS (Quality of Service) assurance of Web Service composition has become crucial and significant challenges.Many research institutions home and abroad have conducted quite a few researches on QoS assurance of service composition, and have made some valuable achievements. However, some key technologies related to QoS assurance of service composition are not well developed. For example, the service reputation measurement, service selection, fault diagnosis, and exception handling. New issues keep emerging.To address the requirements of resource sharing and application integration in open computing environment, some key issues of QoS assurance for service composition have been discussed in this dissertation:Web service reputation measurement based on QoS similarity, Web service selection with uncertain QoS information, fault diagnosis for Web service composition based on Bayesian network, and exception handling for Web service composition using meta-process. The contributions of this dissertation are as follows:1. Reputation is the basic for verity judgement of advertised QoS information and Web service selection. Currently, the research on reputation measurement of web service is mainly based on users’subjective score. Consequently, some characteristics of web service are not considered deliberately. Therefore, a novel Web service reputation measurement model based on the similarity of advertised and factual QoS is proposed in this dissertation. Upper and lower QoS ontology are designed to describe the semantic of QoS and the hierarchical QoS item system. The proposed QoS similarity computing algorithm can support the hierarchical QoS item similarity computing. Subsequently, a statistics-theory-based Web service reputation measurement method is put forward and it uses the QoS similarity to compute Web service reputation. Evolution trend and fluctuation of QoS are considered in reputation measurement with forgetting factor and fluctuating factor. Experiments show that the model is flexible and practical.2. How to dynamically select web services which can best meet the QoS requirements of consumers is very critical to the success of Web services in business domains. Most of current researches use precise QoS information for Web service selection but the uncertainty of QoS information in Web service selection is not well considered. After analyzing the uncertainty of OoS information in Web service selection, the specification method of uncertain QoS advertisement value, QoS requirement value and weight constraints are designed. Then a Web service selection model is built using possibility degree method for interval numbers ranking and TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) method for MADM (Multiple Attribute Decision Making) problems. The experimental results show that the model can effectively select Web service which can meet the requirements of consumers when QoS information is uncertain. Finally, a QoS aggregation method based on aggregation tree for structure Web service composition and a global service selection algorithm is proposed.3. To identify the component services that cause the problem of Web service composition, a fault diagnosis model based on Bayesian network to identify the most likely problematic services in a Web service composition process is proposed. Firstly, the Bayesian network topology construction method and the parameters configuration method are specified in detail. Then a fault diagnosis algorithm based on the Bayesian network is proposed and the algorithm is analyzed. For reducing the collection cost of evidences, we propose an evidence channel selection method based on k-median algorithm and the concept of information entropy and conditional entropy. The experimental simulation show that the model can rule out the root cause of problems in the Web service composition process effectively and efficiently. Compared with the existing research, the uncertainty in fault diagnosis process of service composition and evidence channel selection method are considered in the model.4. For modular, extensible and adaptable exception handling of web service composition, a model using meta-process is proposed. The detail semantic of the model components and their relationship are specified too. The meta-process is a kind of upper control process consists of five meta-activities and the rule in the model is based on event, activity state and case data of base process. When an exceptional event occurs, corresponding meta-process is invoked as controlling and coordinating mechanism for base processes but not only a activity is invoked if some rules are met. Examples analysis show that the proposed model not only can be used for failure recovery, but also can be used for more complex situation such as process modification.5. Based on aforementioned research, a Web service composition reference framework with QoS assurance is proposed. The framework exhibits our understanding for the QoS assurance technology system of composite service and it give a reference computing framework for development of Web service composition support platform and application based on Web service composition.To sum up, we present well-evaluated solutions in this dissertation for some key issues of QoS assurance for service composition. We believe that our contributions make a nice groundwork for future research and engineering on service composition both in theory and practice.更多还原