【作者】 李飞；

【导师】 杨放春；

【作者基本信息】 北京邮电大学 ， 计算机科学与技术， 2008， 博士

【摘要】 Web服务的QoS指的是Web服务的非功能属性,如可靠性,响应时间等等。QoS对于用户使用服务的体验至关重要。在一个有着大量服务的开放服务市场中,服务可能广泛分布于不同的地区,从属于不同的商业组织,如何有效的管理这些服务的QoS,向服务用户提供一个开放,可靠的服务使用环境,就成为了Web服务研究的一个重要的问题。目前的研究中,Web服务的QoS管理方法主要基于集中式管理的思路,集中式管理不能跨越多个管理域,同时也可能面临性能和可扩展性问题。结合P2P计算思想进行分布式的Web服务基础设施研究是近年来Web服务研究的一大趋势,已有了一定程度的发展,但是分布式QoS管理的研究尚属空白。基于对当前QoS管理研究现状和Web服务研究趋势的分析,本文首次对Web服务的分布式QoS管理中的关键问题进行了深入研究,取得如下成果:1.提出了独立的分布式QoS注册方法。QoS具有高度动态性,针对目前分布式服务注册方法并不适用于QoS管理的问题,提出了与服务功能信息分离的分布式QoS注册方法,以及基于此方法的QoS注册系统Q-Peer。利用用户QoS访问的行为特点,Q-Peer设计了高效的查询机制,达到了与集中式QoS注册方式相近的QoS查询性能。2.提出了基于协商的负载均衡方法。为了有效利用分布式QoS注册节点的资源、实现负载均衡,本文研究了Q-Peer中的负载信息特点及其与其他P2P网络的不同,提出了两种有效的数据复制方式,在此基础上又提出了一种新的基于协商的改进方法,以提高负载均衡的效率。实验证明,利用Q-Peer的分布式特性,数据复制明显提高了QoS注册系统的整体利用率。3.提出了组合服务的迭代选择算法。针对当前组合服务选择方法均要求预先知道组合服务的拓扑逻辑和所有备选QoS输入数据,完全不适用于分布式QoS管理的问题,本文提出一种迭代选择算法,可以跨越多个QoS代理节点进行组合服务选择。实验证明,在超过现有算法的算法性能的同时,迭代选择算法达到了很好的选择效果。迭代选择算法算法与现有算法相比还能够自然的支持动态服务选择,并且可以在计算中包含网络因素对组合服务的影响,对于网络敏感的服务也有较大的应用价值。4.提出了事件驱动的分布式QoS监控方法。针对目前QoS监控研究中不能跨管理域监控组合服务,并且对Web服务QoS指标处理灵活性不足的问题,本文提出一种事件驱动的分布式QoS监控方法。该方法基于事件处理规则和事件流模型,以运行时指标为中心,同时提供运行时异常响应和长期指标收集的能力。在此基础上,提出分布式QoS管理环境中的事件路由机制,达到了对跨域组合服务进行灵活监控的目标。最后,提出了基于策略的监控需求描述方法及其向监控事件流自动转化的方法,以支持用户对QoS监控进行个性化配置。5.提出了分布式QoS管理系统。基于以上对分布式QoS管理中关键问题的研究,本文将Q-Peer扩展为分布式QoS管理系统,作为QoS管理能力的承载平台。Q-Peer由大量独立的QoS管理节点构成一个QoS管理的叠加网,采用标准的Web服务协议以及基于角色的接入控制策略,保持了节点间、节点与被管理对象间的松耦合,保证了节点的自治性。Q-Peer将管理能力作为应用,与底层结构相分离,管理功能的扩展不影响Q-Peer的系统结构。更多还原

【Abstract】 Web service QoS or quality of web service is the non-functional attributes of a service, e.g. reliability, response time. QoS is critical to user experience of services. In an open service market, a large number of services may spread in different areas and belong to different business organizations. How to manage QoS in such an environment and provide an open, reliable environment for users to employ services becomes a crucial problem in web service research.In literature, QoS management approaches are based on centralized QoS brokers, which have potential performance and scalability problem, and more importantly, the centralized approaches can not support cross-domain QoS management for web services. In recent years, a major trend in web service research is on distributed infrastructure by leveraging the thoughts of P2P computing, but distributed QoS management has not been substantially explored yet. Based on current status of QoS management and web service research, we address several critical problems in distributed QoS management. The contribution of this dissertation includes:1. Propose an independent distributed QoS registration approach. QoS information is dynamic and everchanging, so currently known distributed service registry works are not suitable to QoS management problem. We propose an independent distributed QoS registration approach where the QoS storage and service functional data storage is separated (The system is called Q-Peer). By analyzing the QoS accessing behavior of users, we propose a highly efficient QoS query approach which achieves query efficiency close to centralized registry.2. Propose a load-balance approach for distributed QoS registry. To utilize the resource of distributed registry, and achieve load-balance, we research the difference on load information between Q-Peer and other P2P network. We propose two effective data replication schemes. Based on the schemes we propose a novel replication scheme by peer negotiation to improve the efficiency in replication. Experimental results show that the replication scheme improves registry utility significantly.3. Propose an iterative selection algorithm for composite service. Known selection approaches requires the information of complete composite service topology and all the QoS of candidate services, which is not applicable in Q-Peer, so we propose an iterative selection algorithm, which can optimize composite service across multiple peers. The experimental results proved that the algorithm achieves excellent selection result and outperforms current selection approaches. The iterative selection algorithm can be applied to dynamic selection naturally. The algorithm includes network QoS as optional factors, which makes the algorithm applicable for network sensitive service.4. Propose an event-driven distributed QoS monitoring approach. To address the problem of known QoS monitors which are not capable of monitoring cross-domain services and not flexible in monitoring complex and personalized QoS, we propose an event-driven distributed QoS monitoring approach. The approach bases on ECA rules and event stream model, to provide run-time QoS exception handling and long-term metric collection. We further propose an event routing mechanism for flexible monitoring cross-domain composite service in distributed QoS management environment. At last, we propose a policy-based requirements description approach to support user configuration for personalized QoS monitoring.5. Propose a distributed QoS management system. Based on our research on the problems in distributed QoS management, we extend Q-Peer registry to a distributed QoS management system as the supporting platform for management capabilities. Q-Peer is an overlay network for QoS management, which is formed by a large number of independent peers. Peers coordinate with each other by standard web service protocols and keep autonomy by role-based access control policy. On Q-Peer, the QoS management capabilities are deployed as applications over access control and communication infrastrusture. The Q-Peer achieves extensibility of management functions as well as management domains.更多还原