【作者】 万虎；

【导师】 费奇；

【作者基本信息】 华中科技大学 ， 系统工程， 2008， 博士

【摘要】 随着人类科学研究及社会实践活动的不断发展,分布式仿真系统越来越复杂。现代大规模分布式仿真对大量计算资源和数据的获取需求越来越高,并且需要使用在物理上处于分布式和隶属于不同组织的资源。新兴的网格技术目标为使用标准、开放的通信协议和接口实现分布式资源的协同管理,提供非平凡的服务质量,可为分布式仿真技术的发展提供良好的资源管理平台。如何结合网格技术对资源管理的优势为分布式仿真的发展提供良好的平台是一个重要的、较新的研究方向。本文围绕如何在网格环境组建分布式仿真涉及到的几个重要问题进行了研究。首先,为了在网格环境中构建分布式仿真应用,本文提出了网格环境中分布式仿真的集成框架GDSF(Grid-based Distributed Simulation Framework)。GDSF的目标是为构建各种领域相关的基于网格的分布式仿真系统提供了一个通用的实现框架。GDSF通过仿真功能模块提供仿真网格中基本的、通用的服务,通过这些模块间的交互,仿真网格平台可为用户提供包括仿真应用与模型的开发、仿真任务的提交、计算资源的选择、仿真应用的监控与管理以及仿真过程与结果数据查询等功能。由于分布式仿真应用的独特性,将其移植到网格环境中执行时,将涉及到较多需要研究解决的问题。本文接下来对在网格环境中开发和执行仿真应用涉及到的三个重要问题进行了研究,问题包括:HLA向网格环境的扩展、仿真服务资源的组织与发现以及仿真应用任务调度中的计算节点选择。1)基于HLA/RTI体系的分布式仿真已得到广泛应用,基于HLA/RTI开发的仿真模块如何融入到网格环境以及如何在网格环境组织基于HLA/RTI的仿真是需要研究的问题。针对上述问题,本文提出了基于网格服务实现HLA/RTI向网格环境扩展的系统结构。系统结构在考虑融入网格技术优势之外,同时考虑到了仿真组件的可重用性、互操作性和兼容性等问题。进一步研究了非基于HLA开发的仿真模块及其他类型服务如何融入到HLA/RTI系统结构中,给出了其通信结构。原型系统实验验证系统结构的可行性。2)仿真服务指以服务形式参与到仿真应用的子模块或者仿真应用运行过程访问的子服务。仿真服务资源是直接参与仿真应用核心资源,对其的有效管理是构建分布式仿真应用的前提和基础,是为仿真用户提供稳定、高效服务的关键。本文随后对GDSF中仿真服务资源管理问题进行了研究:在分析仿真资源的特征的基础上,仿真服务资源的局部集中、整体动态分布式的组织管理结构;给出了基于语义的仿真服务的描述;并根据语义描述给出了基于领域本体的仿真服务匹配算法。3)在仿真应用任务调度之前,如何针对具体仿真用户和应用的需求,优化选择计算节点资源,使仿真应用在满足需求的资源上顺利、高效和稳定的执行是一个需要解决的重要问题。本文最后对仿真应用任务调度过程中的网格计算节点的选择问题进行了研究:分析了基于GDSF仿真网格中仿真应用的常见的通信模式,将仿真应用分为计算密集的序列型仿真和交互通信密集型仿真。随后根据不同类型通信模式仿真应用对通信的需求,利用计算经济学和图论方法分别建立仿真应用任务计算节点选择模型,并设计相应的优化选择算法。相应的实验验证了该算法显著提高了应用执行和资源使用效率。更多还原

【Abstract】 With the continuous development of human scientifical research and social practice activities, distributed simulation system is becoming more and more complex. Modern large scale distributed simulation places strongly requirements on huge amounts of computing and data resources, which will be geographically distributed and subjected to different organizations. As a newly arisen technology, computing Grid is an integrated computing and resource infrastructure that implements distributed resource coordinative operations and guarantees a certain level of QoS to users based on standardized, open and common protocols and interfaces. It provides favorable resource management platform for distributed simulation.To take advantage of computational Grid in distributed resources collaboration and management for distributed simulation technology development is an important, new research direction. This paper focuses on several important subjects involving in establishing distributed simulation application on Grid environment.Firstly, to construct Grid-based distributed simulation system, Grid-based Distributed Simulation Framework (GDSF) is proposed, which aims to providing a general implementation framework for establishing domain specific Grid-based distributed simulation system. Through the interaction of the simulation function modules of GDSF, Grid-based simulation platform provides universal basic services including modeling, simulation application development, simulation task submitting, computing resource scheduling, job monitoring and simulation results inquiry.Because of the particularity of distributed simulation application, to run it on Grid envirionment, many involved problems need to be researched. Then in this paper, three problems including extending HLA to Grid environment, simulation service resource organization and discovery, computing node selection for simulation application tasks scheduling, involving in simulation application development and execution on Grid environment, are studied.1) The HLA-based distributed simulation has been widely used. According to the requirement that integrating simulation federate modules developed based on HLA/RTI into Grid and organizing HLA-based simulation application in Grid environment, a system frame which realizes extending HLA/RTI to Grid environment based on Grid service is proposed. The frame aims to the advantage of Grid technology as well as the reusability and interoperability of simulation modules. And to integrate not-HLA-based simulation modules and other kinds of service into HLA/RTI system, according communication structure is given. The results of experiments of the prototype indicate the feasibility of the frame.2 ) Simulation service includes the sub-simulation-modules that participate in simulation application on form of service and the services that accessed in process simulation application execution. Since simulation service resource is the key resource that directly participates in simulation application, the effective management of it plays a prominent role in running Grid-based simulation application and is important for providing stable effective service to user. Subsequently, this paper studies the problem of simulation service resource management, which including: A local hierarchical-concentration and global dynamically-distributed simulation service resource organization and management model that considering simulation resource characteristic, semantic representation of simulation service, and the semantic-based matching of simulation service based domain ontology.3 ) Before simulation application task scheduling, according to the specific requirement of simulation user and application, in order to execute simulation application efficiently and stably, optimal selection of computing nodes is the key problem. This paper studies the problem of Grid computing node selection on the process of simulation application tasks scheduling. Firstly simulation applications are classified as compute-intensive applications and communication-intensive applications based on analyzing the common communication patterns of the simulation applications running on the Grid. And then according to requirement of different simulation application communication patterns, using computational economics and graph theory, the computing node selection models and optimization algorithms are proposed for the two kinds of simulation application respectively. The corresponding experiments verify that the algorithms significantly improve the resource use efficiency and application execution performance.更多还原