多机群网格中的并行计算实现方法  

Implementation Method for Parallel Computing in Multi-Cluster Grid

作　　者：陈庆奎[1] 崔同云[1] 徐福缘[2] 

机构地区：[1]上海理工大学计算机工程学院,上海200093 [2]上海理工大学管理学院,上海200093

出　　处：《吉林大学学报（信息科学版）》2007年第3期286-291,共6页Journal of Jilin University（Information Science Edition）

基　　金：国家自然科学基金资助项目(60573108);上海市重点学科建设基金资助项目(T0502);上海教委发展基金资助项目(06QZ002;07ZZ92);上海自然科学基金资助项目(04ZR14100)

摘　　要：在由多计算机机群构成的网格环境下,为了实现数据并行型计算,给出了由多计算机机群组成的网格、逻辑计算机机群、数据并行型计算和一系列Agent的定义;利用管理智能体、协同计算智能体、通信智能体以及协同计算组之间的协同计算机制来实现数据并行型计算。根据计算节点的主态、冗态和失效状态,描述了动态资源划分机制和网格计算过程。实践表明,该模型有效地适应了多机群网格环境的异构性、动态性等特性,降低了并行计算的失效率,其上运行的并行计算的加速比呈近线性趋势。该模型适合于基于多机群网格的并行型计算。For implementing DPC （Data Parallel Computing） in muhi-cluster grid that composed of many computer clusters, a series of formal definitions, such as multi-cluster grid, logical computer cluster and three kinds of agents were given. By using a cooperation mechanism among the management Agent, the cooperation compu- ting Agent, the communication agent and the cooperation computing team, the DPC process were studied. Based on the state of computer, such as main state, redundancy state and failure state, dynamic resource partition and grid computing process were described. The experiments show that this method can effectively adapt to the heterogeneous and dynamic characteristic in multi-cluster grid. The speed up of parallel computing by this method is approximately linear. It can be fit for the parallel computing in gird.

关 键 词：多机群网格 多智能体 数据并行型计算 协同机制 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]