【作者】 方雷；

【导师】 刘仁义； 刘南；

【作者基本信息】 浙江大学 ， 地图学与地理信息系统， 2011， 博士

【摘要】 本研究以云计算(Cloud Computing)的关键技术理论为出发点,提出了土地资源信息化管理研究中基于云计算的服务高效处理建模理论框架,解决了空间数据分布式存储策略、空间云服务索引创建与操作、空间数据高效并行操作等关键问题,并通过计算机编程构建了土地资源云平台(Cloud Service Platform of Land Resource,LRCSP)。最后,基于本文的理论框架和建模平台,对土地资源服务的高效处理做了4项实验研究,并展开了深入分析与讨论。具体来说,文本的研究工作主要包括以下几个方面:1、总结云计算的基本理论及应用成果。重点讨论了云计算的基本体系结构,关键技术理论和4个成功的商业云计算平台参考架构。从三个学科的视角出发,总结了云GIS的内涵。以上述两点研究为基础提出一种云GIS的六层体系构架:物理层、虚拟层、数据层、服务组件层、服务层和应用层。着重研究了该架构中的云计算各节点的自动部署策略;并针对平台的前台透明服务需求,从GIS服务的特点出发,提出云GIS服务模型,尤其设计了云GIS服务目录和可供用户进行简单编程的服务接口。针对平台的后台处理需求,提出云GIS平台的高性能并行处理数学模型:参照OGC服务链聚合模式,以原子服务和组合服务的角度分析了功能分解性;同时从矢量数据和栅格数据的数据结构出发,分析了两者的数据可分解性。最后,以上述研究为基础,从面向云计算的土地资源服务特点出发,提出面向云计算的土地资源云平台模型(LRCSP)。2、分析了云计算的关键技术,并以此为基础提出对应的云GIS平台需要解决的关键技术:空间数据的分布式存储策略、虚拟计算节点任务分配模型、基于瓦片的动态地图发布策略以及并行数据库与MapReduce相结合的高效处理模型。在空间数据的分布式存储策略研究中,除了提出基于格网预处理的STRTree的矢量数据并行划分的分布式存储策略和基于四叉树索引的栅格数据的并行划分的分布式存储策略之外,还创新性的提出了基于数据划分的最佳并行策略数学模型。在虚拟计算节点任务分配模型研究中,详细描述了任务分配算法,并在迁移决策阶段提出了计算节点的计算力模型。基于瓦片的动态地图发布策略即解决了一份空间数据经过数据划分并分布存储之后的地图可视化问题,也解决了由于土地数据变更频繁引起的可编辑地图动态更新的问题。在并行数据库与MapReduce相结合的高效处理模型研究中,研究两者优势互补的高效处理模型,并以分地类地物个数统计为例设计了利用MapReduce进行并行统计的算法,为其他类似的并行计算功能提供了借鉴。3、设计实现了原型平台并进行4组对比测试实验。实现并展示了土地资源云平台的3个功能模块:云资源管理模块、土地业务集成子系统和通用客户端。选取大数据量的矢量和栅格数据对土地资源云平台的4项关键技术进行测试。它们是:云存储性能测试中进行栅格数据并行剖分效率对比测试和矢量数据剖分效果对比测试;地图服务浏览性能测试中通过多次加载多计算节点的海量数据对效率进行测试;高效处理性能测试中对多节点的土地数据进行分类统计,用以确定MapReduce的编程模式下的服务效率;虚拟化负载均衡对比测试中将运行虚拟节点上的LRCSP与只安装一个操作系统的普通PC组成的集群系统进行容错、耗能和运行效率的对比测试。研究结果表明,本文选取云计算的基本理论、方法和技术解决土地资源管理问题的路径正确;提出的面向云计算的土地资源服务平台模型展现出高效性、灵活性及扩展性,达到预期目标。作者在土地资源云平台的理论研究及实践作为云GIS应用的一个补充,为后续研究与工作提供了良好的基础。更多还原

【Abstract】 Based on the theory and key technologies of cloud computing, this research firstly proposes a theoretical cloud-computing-based framework for high-performance processing of land resource services. The study also introduces solutions to some key issues such as distributed storage of spatial data, index creation and operation of spatial cloud services and high-performance parallel processing of spatial data. On the basis of the framework, this research establishes a cloud-computing-based application for managing land resource information named Cloud Service Platform of Land Resource (LRCSP). Using this platform, four experiments are performed for testing the efficiency of high-performance processing of land resource cloud services.Specifically, this research mainly includes three aspects as follows:1) A cloud-computing-based framework for managing land resources servicesThis study reviews basic theory and applications of cloud computing, focusing on basic architectures, key technologies and reference architectures of four enterprise cloud computing platforms. From the view of three disciplines, the author discusses definitions of cloud GIS and proposes a six-layer architecture for cloud GIS (physical layer, virtual layer, data source layer, support platform&service component layer and application layer), in which distribution strategy of computing nodes is of great concern. Based on characteristics of GIS services, the author introduces a GIS service model and designs cloud GIS service catalog and service interfaces which allow users to develop some simple programs. The High-performance processing model is a focus of this research. The author analyzes function decomposition in terms of atomic services and composition services, and data decomposition based upon the data structure of vector and raster data. One the basis of the above research, the framework for Cloud Service Platform of Land Resource (LRCSP) is proposed.2) Key technologies to implement LRCSRBased on key technologies of cloud computing, this work presents solutions to four main issues for cloud GIS. For distributed storage of spatial data, this research introduces a Grid-aided and STR-Tree-based partition (GASTRSDP) method for dividing vector data and a quadtree-index-based partition approach for partitioning raster data. In particular, this work puts forwards a data-partition-based algorithm for determining the optimal parallel processing strategy. In the research of job scheduling for virtual computing nodes, the author describes job scheduling algorithms in details and proposes a computational model for computing nodes. To solve the problem of map retrieving after the map has been partitioned and parallel stored, this study chooses a tile-caching-based strategy which can also be used to dynamically update land maps once land data are changed. When using parallel database and MapReduce to improve performance, the author designs a Complementary Model. By using the example of counting the instances of each feature’s type of land-use in a layer, the research presents an algorithm for parallel statistics.3) Implementation of the platform and four experiments to test performance of LRCSP Three functional modules for client, land resources business and cloud resources management are established. Using a large quantity of vector and raster data, four experiments are conducted on cloud storage performance (parallel partitioning vector and raster data), map display performance (loading mass data of multiple computing nodes), high-performance processing (statistics on land use using MapReduce) and virtual load balancing (comparing the performance and consumptions of virtual nodes to those of Cluster System with common PCs that install only one Operation System). The results show that using cloud theory, methods and technologies to solve the problem of land resource management is promising and that the LRCSP is of high performance, flexibility and scalability, which can shed lights on application of cloud GIS.更多还原