节点文献
水科学应用网格的若干关键技术研究
Research on Some Key Technologies of the Water Science Grid
【作者】 张国义;
【导师】 陈国良;
【作者基本信息】 中国科学技术大学 , 计算机应用, 2007, 博士
【摘要】 水科学应用网格是一特定的应用网格,它为水科学计算和水利工程运用提供一个基于网格的问题综合求解与协同计算环境,并为网格应用的生成、配置、运行与性能优化提供一套行之有效的管理机制与操作手段。本文紧紧围绕水科学应用网格的体系结构、网格应用框架、网格应用的性能优化等方面的问题开展了系统与深入的研究,在理论及应用上取得了一些满意的成果,其主要研究内容与贡献如下:(1)水科学应用网格的体系结构研究:以开放网格服务架构(OGSA,Open Grid Service Architecture)为基础,提出了一个符合面向服务计算(SOC,Service-oriented Computing)规范的应用网格体系结构模型;以网格应用框架和网格应用的性能优化为中心,并以服务域和网格应用编程模型为手段,建立了一套关于资源抽象与服务提供、应用生成与配置管理的网格运行机制;以满足水科学计算与工程应用的需求为原则,研制了一个水科学应用网格的原型系统(简称为e-Water)。(2)网格应用框架研究:结合当前的网格与Web计算的研究进展并参照OGSA网格服务框架(Grid Service Framework)等技术规范,提出了网格应用框架的一个元模型(Metamodel);在改进与扩展分布式异构计算的任务/通道(Task/Channel)模型的基础上,提出了一个基于“服务合约创建—端点属性定义—服务动态绑定”的三阶段式网格应用统一编程模型;在参考美国IBM公司的GAF4J(Grid Application Framework for Java)、英国NEReSC中心的WS-GAF(Web Services Grid Application Framework)等研究成果的基础上,为e-Water研制了一个面向服务计算的网格应用框架。(3)网格应用的性能优化方法研究:根据网格资源的动态变化特性,以实现网格及其应用的最佳性能为主要目标,提出了一个基于“感知、规划与控制”的网格性能操纵模型;根据基于PERT(Program Evaluation and Review Technique)的网络规划技术和最小成本加速策略(即所谓的Min-Max算法),提出了一个应用架构与任务调度双向自适应调整的网格应用性能调适(Performance Tuning)方法;采用基于离散事件模拟的建模与仿真技术,在参考国外MicroGrid、SimGrid和GridSim等项研究成果的基础上,研制了一个网格应用的性能仿真器及其关键算法,可针对不同的资源配置和服务绑定策略进行网格及应用总体性能的仿真分析。(4)基于网格的大型流域洪水控制调度系统的研制:设计与实现了一个基于e-Water的大型流域洪水控制调度系统,其中包括河网水动力模型的并行直接求解、多设定洪水控制调方案的并行累进增量克隆仿真(PPISC,Parallel and Progressive Incremental Simulation Cloning)等关键算法;根据各模块的计算特点,分析了系统的资源需求,提出了系统的总体部署方案;结合网格应用的性能调适方法,建立了该系统的性能模型,并针对不同的资源配置和服务部署策略进行了系统总体性能的对比仿真;以1996、2003年夏季的淮河流域洪水控制调度案例为实验数据对系统进行了实际的运行测试。理论分析与测试结果均表明,采用本文所提出的e-Water网格关键技术,不但使经优化后的PPISC算法具有近似线性的加速比,而且使系统的计算精度与总体性能得到了较大幅度的提高。
【Abstract】 Water science grid is a kind of application grid, which was designed to provide a comprehensive grid-based problem solving and cooperative work environment for the domain of water science and hydraulic engineering, and to provide an effective management mechanism for the integration, configuration, execution and performance optimization of domain-specific applications. This paper researches some key technologies on the grid architecture, grid application framework and methodologies for performance optimization of grid-based applications. The main work and contributions are described as following:(1) The research of the architecture of water science grids. First, based upon the Open Grid Service Architecture (OGSA), an architectural model for water science grids that conforms to the specification of service-oriented computing (SOC) was proposed. Second, to solve problems about the resource abstraction, service provision, system integration and execution management, a grid execution management mechanism coupling the grid application framework and performance optimization methodologies was set up. And last, to maximum satisfy the requirements of the domain-specific applications, a prototype of the water science grid, named e-Water in this paper, was realized.(2) The research of grid application framework. First, by using state-of-the-art grid and Web computing technologies, and by referencing the OGSA grid service framework, a meta-model of grid application framework (GAF) was proposed. Second, by improving and extending the task/channel programming model for distributed heterogeneous computing, a three stages unified programming model based upon "Service contracts construction-End points properties definition-Dynamic service binding" was proposed. And last, by referencing IBM’s grid application framework for java (GAF4J) and NEReSC’s Web Service grid application framework (WS-GAF), a SOC based grid application framework for e-Water was realized.(3) The research of methodologies for optimizing the performance of grid applications. First, to realize objectives of best performance both for grid and grid applications, a performance manipulating model based on "Aware, Planning and Control" was proposed. Second, according to techniques of the PERT net and the Min-max speedup strategy, an approach to tune the performance of grid-based applications, in which a mechanism of bi-directional adjusting the application architecture and the task dispatching were employed, was proposed. And last, by using technologies of discrete event modeling based simulation, and by referencing famous grid simulation system in the world such as MicroGrid, SimGrid, GridSim and etc., a simulator and some key algorithms for e-Water was realized, which can model the performance both of grid and grid application in different scenarios of resource configuration and service binding.(4) Case study. First of all, based upon the e-Water grid, a flood mitigation operation system for large-scale river basins was realized, including some key algorithms such as the directed parallel solver for hydrodynamic models, and algorithms for parallel and progressively incremental simulation cloning (PPISC) of alternative scenarios of flood mitigation operations. Second, based on computational features of each module in the system, the requirements of grid resource were analyzed, and deployment scenarios of the system were discussed. Third, based on performance tuning approaches proposed in the front of this paper, a performance model for the system was set up, and different scenarios of resource configuration and service deployment were simulated. And last, the system was tested by using two real flood mitigation operation cases in 1996, 2003. Theoretical analysis and test results both manifest that, by using the key technologies proposed in this paper, the optimized PPISC has approximately linear speedup, the precision of computations was increased and the whole performance of the system was improved.
【Key words】 water science grid; grid application framework; performance optimization; case study;