【作者】 祝希路；

【导师】 王柏；

【作者基本信息】 北京邮电大学 ， 计算机科学与技术， 2011， 博士

【摘要】 随着互联网技术的不断发展,Web服务技术逐渐成为当前的一大研究热点。在Web服务技术众多研究中,Web服务的可信性已成为国际上致力于要解决的重要课题。对于部署在分布、开放、多变、难控的网络环境下的Web服务,保证其能够正确、安全和有效的为用户提供服务显得尤为重要。因此,进行高可信的Web服务研究具有一定的战略性、基础性和前瞻性,对我国软件业发展具有重大意义。随着Web服务技术的不断发展,各种各样的应用也随之出现。因此,服务质量(Quality of Service, QoS)逐渐成为Web服务中的一个关键技术点,而通过对QoS的评价可判断Web服务的可信性。在目前的研究中,基于QoS的可信Web服务评价多集中在可信度量、QoS管理等方面。虽取得了一定的成果,但在某些问题上仍然缺乏深入的研究。基于Web服务的研究趋势和对Web服务进行可信评价问题的研究现状,本文从以下几个角度对基于QoS的可信Web服务问题进行了深入的研究,取得了以下研究成果：1)针对当前分布式QoS代理难于有效支持QoS的范围查询等问题,提出一种基于QoS的分布式Web服务索引框架。该框架由两层结构的分布式系统构成,其中第一层运用Chord网络对具有相同或类似功能的Web服务进行分类管理,第二层主要用于管理领域相关的Web服务的QoS。为构建用于分布式管理的QoS多维空间,将分类管理的QoS进行归一化,并引入了基于KD-Tree的多维数据管理方式。KD-Tree实现了QoS的范围查询,通过范围查询的方式,可筛选掉大量不满足非功能需求的Web服务。为保证分布式系统的查询效率,提出了基于QoS-Tree的管理方式。QoS-Tree是通过占优关系而构建的,其不仅实现了KD-Tree中涉及到的范围查询功能,而且通过QoS-Tree中占优关系可以较容易的找到满足用户需求的QoS。另外,为避免分布式QoS-Tree中查询负载不均衡,一种负载均衡机制被引入到系统中。实验表明所提出的两种分布式树型结构均能有效支持QoS的范围查询,且QoS-Tree的查询效率要优寸KD-Tree。2)针对开放环境对Web服务中某些QoS指标造成的影响,提出了基于区间型QoS和概率区间型QoS的度量方式以及相应的服务选择算法。区间型QoS的度量方式是通过比较QoS区间上的概率密度来获取QoS中某些指标受动态环境影响的程度。这种方式正确性虽高,但其比较方式只适合于单个或少量的服务选择。概率区间型QoS的度量方式主要是通过计算QoS概率密度区间的置信上、下界与其均值的偏差来描述动态环境对Web服务的影响,因此,最优组合服务的选择可采用均值的加权和最大以及偏差最小来度量。将该度量方式转化为单目标组合优化问题,提出了一种基于社团划分的改进遗传算法。算法通过社团划分方式维护种群的多样性,并通过优良模式的叠加以提高种群收敛速度。仿真实验表明两种度量方法均能有效克服动态环境对服务选择的影响,且基于区间型QoS的度量方式在正确率方面要比基于概率区间型QoS略高。但概率密度的度量方式更适合于采用遗传算法进行全局寻优。实验表明,改进遗传算法提高了服务选择的效率。3)为提高Web服务选择过程中服务之间的合作可信,提出了一种Web服务合作声誉模型。为描述这种合作信誉,构建了Web服务合作网(Web Service Collaboration Network, WSCN)。在合作网中,邻居更新机制能淘汰虚假服务,保证候选服务的可信性。根据合作网结构,任何Web服务都包含两种关系,调用关系和合作关系。因此,Web服务合作声誉可通过这两种关系进行度量。该声誉模型包括两类指标,调用声誉是通过在WSCN的社区结构中选择适当推荐者而计算得到；被调用声誉指标是通过Web服务之间的调用频率来评价。基于WSCN,提出一种Web可信服务选择算法。实验表明WSCN保证了服务选择的可信性,声誉模型能有效的组织Web服务之间最可信的服务进行组合,同时合作关系能有效地加速服务选择的过程。4)为解决海量Web服务带来的服务信息维护开销,提出一种基于云计算的Web服务管理平台。该平台分为四层,最底层由Hadoop分布式文件系统构成,用于支撑整个云环境。HBase支撑Web服务信息注册层,主要用于管理从Web服务中获取的功能和非功能属性。此外,为进一步提高Web服务选择效率,HBase表中还负责维护了Web服务之间的合作关系以及本体树和QoS-Tree。平台中的核心层使用的是MapReduce技术,该层主要是为满足Web服务信息的有效存储,提出了基于MapReduce的QoS数据提取、转换、加载操作和数据挖掘算法。在复杂的Web服务选择方面,通过MapReduce并行搜索所有可能的路径,并获取其中的最优解决方案。实验结果表明所提出的框架更适合海量Web服务的管理。更多还原

【Abstract】 With network technology’s development, many researchers have done lots of job in web service field. In many research areas of web service, the trustworthiness becomes the significant research subject. Due to the open, distributed, uncertain and uncontrollable network environment which web service deploys, providing accurate, secure and effective service for customers becomes especially important. Thus, the research on high-trustworthy web service is a prospective, basic and strategic task, it plays a significant role in the software development.With web service technology’s development, various applications based on it have emerged. The QoS research increasingly becomes a critical problem, and the trustworthiness of web service can be assessed by QoS. In recent years, the approaches to evaluate web service’s trust based on QoS concentrate in trust evaluation, QoS management, and so on. However, some areas have not been well researched or even considered. Based on current web service research and status of trust evaluation based on QoS, we address some critical problems on trustworthy web service based on QoS.The contribution of this dissertation includes:1) To solve the problem that distributed QoS broker cannot support range query, a distributed web service index framework based on QoS is proposed. The framework consists of two kinds of structured distributed system, one is chord overlay network which is assigned to classify the web services according to the similarity of the functional properties, the other manages QoS of domain-related web service. To construct a multi-dimension QoS space, the classified QoS is normalized, and the distributed tree structure KD-Tree is introduced. In KD-Tree, the range query method is realized, by using this method, all satisfied QoS can be obtained. To improve the query effectiveness, another distributed tree structure QoS-Tree is proposed, and the range query method is also supported by QoS-Tree. Due to the dominated relationship in QoS-Tree, the query efficiency can be guaranteed. Additionally, to avoid the imbalance of query load in the distributed system, a load balance strategy is introduced which can balance workload efficiently. Experimental results validate the effectiveness of the proposed distributed tree structure approaches for complex query, and query cost of QoS-Tree is better than that of KD-Tree.2) To solve the effect which open environment imposes on some QoS, a web service selection algorithm based on interval and probability interval QoS are proposed. The interval QoS measurement can effectively assess the uncertainty by comparing the probability density in QoS interval. Although the accuracy can be guaranteed, this method can only apply for a few web services to be combined. The effect described by probability interval can be evaluated by using the difference between the mean of probability density and the upper, lower bounds of probability density. Thus, the composite web service with probability interval QoS global optimization can be evaluated with the maximum weighted sum of mean and the minimum difference, and this measurement is transformed into combinatorial optimization, and a genetic algorithm is developed based on the community detection. The population diversity can be maintained by the multi-community structure, and excellent schema combination accelerates the population’s convergence. The experimental results show that these two methods contribute to reduce the influence of dynamic environment. The accuracy measured by interval QoS is slightly higher than that of probability interval QoS, however, the genetic algorithm can apply to service selection based on probability interval to search for global optimum. The experiment of the improved GA shows that the selection efficiency based on probability interval QoS is enhanced.3) To improve the collaboration trust between web services during selection process, a novel reputation model called collaboration reputation is proposed. To describe this reputation model, a web service collaboration network is constructed. In WSCN, the false web service is eliminated by using the neighbor update mechanism, and the credibility of web service selection can be guaranteed. According to the collaboration network, any web service has two types of relationship, invoking and invoked relationship. Therefore, the collaboration reputation can be assessed by these relationships. It includes two metrics; one called invoking reputation is computed by the proper recommender, which is selected in WSCN’s community, the other is assessed by the invoked web service. In addition, the web service selection based on WSCN is designed. With the collaboration reputation guaranteed, the selection efficiency and the solution’s credibility are both increased. Experimental results confirm that WSCN ensures credibility of web service selection, our reputation model has better ability to organize services to make trustworthy service combination, moreover, the selection process can be effectively sped up by the collaboration relationship among web services.4) To address the maintenance cost which large-scale of web service impose on, web service management platform based on cloud computing is proposed. This platform is consisted of four layers. The bottom layer is based on Hadoop Distributed File System, which functions as a foundation for supporting the upper layers. In information registration layer, HBase is assigned to manage the functional and non-functional properties extracted from web service advertisement. To accurately and quickly select the satisfied web service, the web service collaboration table and interface matching table are designed according to HBase. In addition, the ontology tree and QoS-Tree table are also designed for further enhancing the performance of functional and non-functional property retrieval. The core layer uses MapReduce technique, and the process of extraction, transformation, load and data mining algorithm based on MapReduce are proposed to satisfy the efficient web service management. To satisfy the user’s complex web service selection, the searching optimal solution in all possible paths which depends on MapReduce is proposed. The results of the extensive experiment show that this framework is more suitable for large-scale web service management, and the proposed algorithm provides better performance.更多还原