节点文献

基于网格计算的大规模分布式动态虚拟环境仿真研究

Research on Large Scale Distributed DynamicVirtual Environment Simulation Based on Computing Grid

【作者】 蒋从锋

【导师】 王乘;

【作者基本信息】 华中科技大学 , 水利水电工程, 2007, 博士

【摘要】 大规模虚拟环境及地形可视化仿真在流域水文气象分析、洪水演进模拟、防灾减灾、城市景观规划、虚拟旅游、在线角色游戏等领域越来越得到重视.卫星遥感技术的发展,使得获取高分辨率的地理空间数据成为可能.而计算机有限的处理能力与大规模虚拟环境仿真,尤其是海量地形数据实时显示与交互的矛盾是影响大规模虚拟环境仿真与地形可视化应用的一个主要障碍.另外,随着网络技术的进步与普及,实现远程在线协同仿真,将是大规模虚拟环境与地形可视化仿真的发展与应用方向.网格计算技术被认为是下一代的互联网,利用网格计算技术,可以将互联网上的各种资源(超级计算机、大规模存储设施、个人计算机、各种传感器、软件系统和各种外部设备等),整合成一个类似电力网的巨大“计算池”,将各种计算资源虚拟为一台“虚拟超级计算机”,解决大规模虚拟环境与地形可视化仿真中对计算能力的巨大需求,并可以实现远程多单位多主机协同在线仿真.基于以上需求,本文提出了基于网格计算的大规模虚拟环境仿真系统层次化体系结构.在整个体系结构中,处于系统最底层的是节点层,向上依次是通讯层、数据层、计算层、管理层和应用层,最上层是网格门户层.在此体系结构基础上,提出了以下关键算法和模型:(1)网格环境下大规模分布式动态虚拟环境仿真系统的安全与容错任务调度算法SAFTS(Security Aware and Fault-Tolerant Scheduling). SAFTS算法对用户仿真任务的安全需求和可用资源的信任等级进行匹配,在系统安全等级较低并且网络和主机可能失效的网格环境中进行容错任务调度.根据网格系统的安全等级,自适应调整任务备份数,并对失败的任务重新调度.本研究使用模糊推理来确定任务的备份数.仿真结果表明,该算法可以有效提高不安全网格环境下的任务调度成功率,具有很好的容错性和可扩展性.(2)一种改进的基于视点相关的大规模地形层次细节(LOD)显示算法VMLOD(View-dependent based modified Level of Details). VMLOD算法根据视点位置,建立连续的层次细节模型,并根据视截体的投影来对三角形进行裁减,避免了大规模地形绘制时裂缝的出现,并加快了地形绘制效率.(3)网格环境下大规模虚拟环境与地形仿真的海量数据管理模型.本文提出的海量数据管理三层体系结构,支持海量数据统一存储和管理、拓扑关系管理、元数据管理和数据并发处理,提高了大规模地形数据的访问效率、数据一致性和安全性.基于上述研究成果,本研究建立了一个用于大规模虚拟环境与地形仿真的硬件平台和软件平台,主要包括:八节点的局域网网格平台、网络通信系统、计算系统、数据库系统、仿真可视化终端、主控系统及网格操作系统,并编程实现了一个软件原型系统GLTVS(Grid-based Large-scale Terrain Visualization Simulation).在GLTVS中,处于系统最底层的是各种架构的网格计算节点,是整个仿真系统的计算能力提供者,它包括各种同构、异构以及能力不同的计算资源.高性能通信系统将各个网格节点进行互联,保证整个虚拟环境与地形仿真系统各个子节点、子任务间信息的透明传输.在高性能通信系统基础上,是高可靠性和高可用性的数据系统和网格存储系统,该系统不仅可以存储上述信息,还可以在不同节点间进行可靠的文件及数据传输,包括数据备份和恢复.在通信系统之上,是基于网格平台的虚拟环境与地形仿真系统的通用功能模块,包括用户认证系统、任务管理与调度系统和负荷平衡系统.在通用功能模块之上,是虚拟环境与地形仿真系统的应用模块集合,包括环境显示系统、仿真策略与专家知识库及其它专用系统等.GLTVS系统以Web页面作为对外提供服务的统一界面,具有平台无关性、高安全性和高易用性.

【Abstract】 Large-scale virtual environment simulation and terrain visualization have been regarded as a key technique in the field such as watershed hydrological analysis, climate analysis and simulation, flood routing simulation, disaster prevention and alleviation, urban planning, virtual traveling, online games, and so on. The advances in Remote Sensing (RS) make it possible to acquire geographical and spatial data with higher resolution.Unfortunately, large-scale terrain visualization is not widely used and the performance is not satisfied due to the processing power limitations. In the future, remote online cooperative visualization and simulation will become true with the advances in computer networks and its wide usage.Grid computing, regarded as the Next Generation Internet, can help aggregate the resources on the Web, such as supercomputers, large-scale storage infrastructures, PCs, sensors, software, and computer peripherals, to a“Virtual Supercomputer”to break the processing power limitations in large-scale terrain visualization and remote cooperative simulations.In this thesis, a hierarchical architecture of grid computing based large scale distributed dynamic terrain visualization and simulation is proposed.The architecture is composed of grid sites, communication and authentication protocols, individual resource management, global resource management, simulation applications, and grid portal, in a view from bottom to top.Based on the architecture proposed above, some key algorithms and system models were proposed, such as:(1) Security Aware and Fault-Tolerant Scheduling (SAFTS) algorithm based on adaptive job replications. In risky and failure-prone grids, SAFTS is robust due to its adaptive job replications and rescheduling mechanism. Simulation results generated from SimGrid package show that the performance of SAFTS is better than non-security-aware scheduling algorithms based on fixed-number job replications and SAFTS is fault tolerant and scalable.(2) View-dependent based modified Level of Details (LOD) algorithm (VMLOD) for large-scale terrain visualization. In VMLOD, continuous LOD model is constructed according to the view position. Simulation results show that VMLOD is faster than conventional LOD algorithms and it can draw the terrain without slits.(3) A hierarchical architecture of massive data management. The architecture is composed of data storage nodes, data transferring and processing system, and computational facilities. The architecture supports global massive data management, topology management, meta-data management, and data concurrency. Experiment results show that the data management architecture here developed is highly efficient, reliable and robust. After comprehensive theory study of above mentioned key problems, a prototype system for Grid-based Large-scale Terrain Visualization Simulation(GLTVS) is constructed based on the above research results, which mainly consists of hardware and support software, including networking communication sub-system, computing sub-system, database sub-system, visualization terminals, control console terminal, and grid Operating System.In GLTVS, processing power is provided by the grid sites and the data are transferred transparently through grid sites and tasks. The grid storage subsystem provides high data availability and reliability, including file and data transferring, data replication and recovery. Computing sub-system includes various homogenous and heterogeneous computing resources with different computing powers.The common functional modules aggregate the applications such as environment visualization, simulation strategies and some other specific subsystems.In GLTVS, all the grid services are provided through a grid portal which makes the service easily accessible, safe, high usability, and can be easily manipulated by nonprofessional users. Tests of GLTVS were carried out on LAN (Local Area Network) gird computing platform. Results show that the hardware and software architecture of GLTVS we developed is efficient, reliable and robust. The response time of database server is within 1.5 seconds and can satisfy real time operation.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络