【作者】 李嘉；

【导师】 王乘；

【作者基本信息】 华中科技大学 ， 系统分析与集成， 2005， 硕士

【摘要】 大规模分布式仿真所涉及的数据空间非常巨大,需要耗费大量的时间,并且对计算环境也有较高的要求。在如此巨大的分布式环境中对资源进行管理是一项十分复杂的任务。将网格技术应用于大规模分布式仿真中,能有效地解决上述问题。为了解决HLA 仿真系统在资源调度上的问题,该文在OGSA 的基础上,设计了服务于分布式仿真系统的网格资源管理系统框架。结合系统框架和分布式仿真的运行特征,对标准仿真应用层进行了改造,并详细设计了资源管理系统中各功能模块的实现和模块间的工作流程。对等(Peer to Peer)技术将点到点(Point to Point)的通信从主从式(Client/Server)结构进一步转变为对等体间的连接。该文将P2P 技术引入网格资管理系统的实现中,剔出了系统中的集权节点和中心节点,将相应的功能转入各个仿真计算节点上,实现了仿真节点的自治运算。系统中不再存在必不可少的全局节点,只会根据仿真任务动态地生成局部管理节点,有效解决了分布式仿真系统中的单点实效问题。该文最后设计并实现了网格资源管理系统对仿真节点的负载检测办法和调度算法,并通过实例进行检验。结果表明设计实现的网格资源管理系统能够有效稳定地利用节点资源,具有良好的负载平衡特性。更多还原

【Abstract】 Many kinds of resources in large and distributed environment related to computing environment are required by distributed simulation, and management of these resources is a complex task, which can be accomplished by combining it with the techniques of grid. In order to solve some problems in current High Level Architecture, such as lack of intelligent and flexible mechanism in managing resource, on the basis of OGSA, this paper designs grid resource management framework serving distributed simulation. According to the run-time features, application layer of HLA is made corresponding transform; the working process between every function modular and these modules is designed in detail. Peer to Peer computing change the point-to-point communication from the client/server architecture to the interconnection of peers. The paper makes use of P2P technology to implement the grid resource administrative system, from which all the central management nodes are removed, and their corresponding function modules are transferred into each simulation node. Thus all the simulation nodes get the power of autonomy computation. System has essential overall node no longer, and generates local temporary management node according to needs for every task. Single-failure problem in distributed simulation system is solved effectively. In this paper, load-measured method and scheduling algorithm to manage simulation computation node is finally designed and implemented. And an experiment is utilized to test the system performance. As a result, show that system can manage simulation resource efficiently and steadily, furthermore has good load balanced property.更多还原