节点文献
非集中式空间服务管理与组合技术研究
Research on the Decentralized Management and Composition Techniques of Geospatial Services
【作者】 左怀玉;
【导师】 景宁;
【作者基本信息】 国防科学技术大学 , 信息与通信工程, 2008, 博士
【摘要】 随着Internet技术和Web技术的发展和大规模普及应用,地理信息系统(Geographic Information System,GIS)网络化逐渐成为一种发展趋势,各种空间信息设施逐渐成为现代社会重要的基础设施,越来越多的空间数据资源、空间计算资源以及空间应用资源依托互联网已经成为公共获取和访问的网络资源,推动传统意义下的空间信息发布逐渐转移到基于Internet的开放性分布式计算基础设施。近年来,随着Web服务技术的出现与推广,面向服务架构的空间信息共享逐渐成为主流范型。空间服务相关技术在国内外得到很多研究团体的关注,从空间服务的创建到空间服务的使用已经积累了大量成熟经验、技术乃至标准和规范。随着相关研究技术的不断深入,能否在面向空间服务的应用体系下进一步增强空间服务使用效率和空间服务管理扩展性成为了当前空间服务技术的一个研究热点。直接将空间服务部署在互联网的方式存在空间服务宿主服务器之间缺少主动协同能力的缺点,因而在空间服务管理(如空间服务的部署、发布、发现以及选择等)上缺少可扩展性,在空间服务组合上缺少提升组合系统性能的能力。随着空间服务实例的增多,上述问题将面临更为巨大的挑战。本文主要针对上述空间服务管理和组合面临的问题进行研究,提出了将空间服务部署到一种基于对等网络的Geo-P2P重叠网络上进行非集中式管理和组合的研究思路。首先提出并研究一种适合于空间服务管理与组合的Geo-P2P网络基础设施,其后针对Geo-P2P网络中的空间服务发现和空间服务组合开展研究。在此过程里,着重研究了Geo-P2P上的网络拓扑、空间服务查询模式、空间服务查询实现、空间服务组合建模以及空间服务组合执行等关键技术。其中,Geo-P2P网络拓扑用来提供空间服务的非集中式高效查询能力并增强整个Geo-P2P的扩展性;空间服务查询模式用来降低查询交互的复杂性;空间服务查询实现主要解决非集中式空间服务查询的具体实现;空间服务组合建模用来解决如何表达非集中式空间服务组合过程;空间服务组合执行则用来解决集中式组合模型的分解以及非集中式组合模型的执行等问题。本文主要工作包括:1.非集中式空间服务的管理与组合框架研究:基于P2P技术提出了一种Geo-P2P网络,通过设计满足空间服务消费者、空间服务提供者、空间服务查询者以及空间服务组合者等多种角色的对等网络节点Geo-Peer来增强空间服务支持环境扩展性和组合性能的能力。从系统结构、管理模型以及组合模型等不同方面研究Geo-P2P网络,为开展非集中式空间服务管理与组合技术研究提供基础性设施支撑。2. Geo-P2P网络拓扑结构研究:通过改进超级节点的网络拓扑结构提出一种基于多层稀疏拓扑的混合网络拓扑作为Geo-P2P的拓扑基础模型。对空间服务查询过程分析后建立了查询代价计算模型,从一般意义上研究了超级节点与普通节点之间的最佳比率,并提出了一种Geo-P2P网络拓扑模型的动态维护策略,实验验证了Geo-P2P网络拓扑结构及其动态维护策略的有效性。3. Geo-P2P网络上的空间服务查询模式研究:提出一种基于信息检索模型的空间服务查询模式,通过将空间服务元数据分离为非空间属性和空间属性来分别建立相似性度量,其中,在非空间属性的相似性度量中,沿用了信息检索技术中矢量模型的相似性度量;在空间属性的相似性度量中,则提出一种包含拓扑、距离以及方向的相似性度量;并用两者之间的加权计算空间服务查询条件和空间服务实体之间的匹配关系。4. Geo-P2P网络上的空间服务查询实现研究:提出一种在对等节点间聚簇成组的空间服务查询和一种基于分布式CAN-QTree索引的空间服务查询。前者基于传统P2P网络的内容聚合优化策略,根据每个对等节点上的空间服务相关性维护一种有利于指导查询的空间服务语义组;后者能够提供有保障的空间服务查询性能,它通过在对等节点之间维护结构化的P2P网络拓扑结构CAN,用QuadTree技术来索引空间服务的地理范围,并用CAN的逻辑空间和地理范围的坐标空间之间逻辑映射形成分布式索引CAN-QTree。CAN-QTree能够提供O(n1/2)的空间服务查询性能。上述两种不同的空间服务查询实现在Geo-P2P网络上的有效性均通过实验得到验证。5.非集中式空间服务组合建模技术研究:针对空间服务组合行为中的全局流程规约、局部流程规约以及全局流程规约与局部流程规约之间的关系,分别提出了全局流程模型、局部流程模型以及子流程模型。在全局流程模型里,分别用活动、控制流和数据流对空间服务组合的步骤、活动之间次序以及活动之间的消息流量和流向进行建模,并通过关键路径概念来定量评价全局流程模型的优劣。在局部流程模型里,增加了协同活动来建模控制流和数据流在多个子流程模型间的跨越边界行为。6. Geo-P2P网络上的空间服务组合执行技术研究:提出一种全局流程模型的间接分解算法,首先将全局流程模型转化为关键路径最短的流程模型,然后再基于消息传输代价最小原则对全局流程模型进行分解。实验表明,与传统的直接分解算法相比,在不同规模的组合消息流量和网络带宽情况下,间接分解算法均优于直接分解算法;而在消息流量规模大或网络带宽有限的情况下,基于Geo-P2P网络的空间服务组合执行明显优于集中式组合机制。7.基于论文研究成果构建了Geo-P2P实验原型系统,重点通过对网络节点Geo-Peer的接口设计,使其能够充当空间服务消费者、空间服务提供者、空间服务查询者以及空间服务组合者等多种角色,并基于上述接口的APIs实现了对前面章节研究成果的综合应用与验证,验证了论文工作的有效性和正确性。
【Abstract】 With large-scale applications of Internet technology and Web technology, Geographical information system network has become an inevitable trend of development. Various kinds of geospatial information system have gradually become important fundamental infrastructures. More and more geospatial data resources, geospatial computing resources and geospatial applications relying on the Internet have become public accessible network resources. All these have promoted the traditional sense of released geospatial information to shift to the open Internet-based distributed computing infrastructure gradually. In recent years, with the maturation of Web services technology and their promotion, service-oriented architecture for geospatial information sharing has become a mainstream paradigm. Geospatial services have been concerned by many domestic and international research groups. From the creation of geospatial services to the use of geospatial services, we have accumulated a lot of mature experience, technology, standards and norms. As technology continues to be studied in depth, we need to further consolidate and promote geospatial service technology by enhancing the efficient use of geospatial service.Hosts of geospatial services can not collaborate actively with each other if we deploy these services directly on the Internet. Such direct deployment also makes geospatial service management (such as geospatial service deployment, publish, discovery and selection) less scalability and depresses the ability of composition system to improve performance. With more and more geospatial services deployed, such questions will be faced with more challenge. In this dissertation, we present a different idea which uses a decentralized geo-p2p overlap network to manage and compose geospatial services. In our methods, we first bring forward a decentralized network infrastructure based on p2p technologies, and then focus on the management and composition for geospatial services with this infrastructure. During our studies, we investigate the technologies of the geo-p2p network topology, geospatial service query model and its implementation on the geo-p2p, modeling and execution of geospatial service composition. Our main contributions are:1. We have presented a decentralized framework for management and composition of geospatial services. In our framework, we bring forward a geo-p2p network based on the traditional p2p technology. We also design geo-peers to construct such decentralized network. All these geo-peers can act as geospatial service consumers, providers, discoverers and composers synchronously. With these active geo-peers, we can enhance the scalability and composition performance of geo-p2p networks. We study different levels of geo-p2p networks from system architecture, geospatial service management model and geospatial service composition model to support further research.2. We have studied geo-p2p network topology and present a hybrid network topology based on the multi-layer sparse topology as well as its dynamic maintenance algorithm. We first establish a geospatial service query cost model by modeling the service discovery process, and then compute an optimal ratio between super geo-peers and basic geo-peers from a general sense of geospatial service discovery. Finally, we give a dynamic maintenance algorithm for the geo-p2p network topology. Experimental results support well with our theoretical conclusions.3. We have probed into query models for geospatial services on geo-p2p networks and present a new query model based on information retrieval technology. In the query model, we propose a non-spatial similarity function and a spatial similarity function to evaluate the degree similarity between a query and a geospatial service partially. In the non-spatial similarity function, both a geospatial service and a query are represented as vectors as those in information retrieval. The vector model proposed can evaluate the degree of similarity of a service and a query as the correlation between those two vectors non-spatially. In the spatial similarity function, both a geospatial service and a query are represented as rectangles. The topology measure, distance measure and direction measure between those two rectangles can be used to evaluate the degree of similarity between the service and the query spatially. Therefore, we can use the two weighted functions ultimately to compute the degree of similarity between each geospatial services stored in the geo-p2p networks and the user query.4. We have implemented two kinds of geospatial service query on geo-p2p networks. The first implementation is to cluster geo-peers into different geospatial semantic groups according to the correlation between geo-peers. This idea is coming from an optimization mechanism called contents aggregation for shared-resources discovery over traditional p2p networks. The query can walk between those groups biasedly, so it can get a better performance in those heuristic groups. The other implementation is to maintain a distributed index called CAN-QTree for geospatial service query over geo-p2p networks. In such method, all geo-peers to be queried interrelate with each other using a structured topological network called CAN, and the metadata of their geospatial services are indexed by a Quadtree data structure. With a deliberate mapping between the zones of the CAN and the underlying space of the Quadtree, we can maintain a distributed index call CAN-QTree on geo-p2p networks to provide a O(n1/2) search performance. All those two methods are verified by our experiments in this dissertation.5. We have explored modeling decentralized geospatial service composition and present global process model, local process model and sub-process model to represent overall geospatial service composition, corresponding geospatial service composition of each geo-peer on geo-p2p network and their relationship. In the global process model, we use activities, control-flows and data-flows to model steps of geospatial service composition, sequences between all activities and messages. We also evaluate quantitatively those global process models through a critical path concept. In local process model, we add collaborative activities to model the control-flows or data-flows which span different local processes.6. We have investigated decentralized executions for geospatial service composition on geo-p2p networks. We present an indirect algorithm for decomposing a global process into different local processes based on best process concept. In the algorithm, we first turn a global process into a best process without losing any functions, and then decompose the best process into several local processes on the principle of minimizing messages transmission between those local processes after decomposition. Our simulations show that the algorithm is better than the traditional direct decomposition algorithm and it is also superior to the implementations of centralized execution for geospatial service composition in the circumstances with large messages or limited network bandwidth.7. We finally design and implement an experimental prototype system according to the research presented above. In our system, we design different interfaces for geo-peers to act as geospatial service consumers, geospatial service providers, geospatial service discoverers and geospatial service composers synchronously. Our experimental prototype system shows that our achievements are effective and practical.