计算网格中多编程环境的研究与实现  

Study of Multi-Programming-Environment in Campus-wide Computa tional Grid

作　　者：陈亚玲[1] 桂小林[1] 王庆江[1] 钱德沛[1] 

机构地区：[1]西安交通大学电信学院新型计算机研究所,西安710049

出　　处：《计算机工程与应用》2004年第17期112-114,123,共4页Computer Engineering and Applications

基　　金：国家自然科学基金项目(编号:60273085);国家863高技术发展研究计划项目(编号:2001AA111081)资助

摘　　要：计算网格(也称元计算系统)聚集地理上分散的资源进行大型的分布式高性能计算。PVM和MPI是广泛使用的并行编程环境,它们需要作为并行计算的基本构建而集成到元计算系统中去。论文针对元计算资源的动态性、分布性、性能多变性和结点异构性等特点,实现了一个自适应的、一体化的多编程环境。论文论述了该多编程环境的体系结构,并利用代理技术实现远程编译、发现资源、屏蔽异构和优化调度。Computational Grid(i.e.Metacomputing system)collects geographically distributed resources for large-scale distributed high performance computing.PVM and MPI ,which are popular parallel programming environments,will be integrated into the metacomputing system as the infrastructure of parallel computing.The resources of metasystem are characterized by dynamic,distribution,performance variety and heterogeneity.For adapting to these characters,a Grid-Aware uniform multi-programming-environment is presented in this paper.The architecture of the multi-programming-environment is given,besides,the key technology called'agents'is presented for remote compiling,exploring resource,masking heterogeneity and optimizing scheduling.

关 键 词：多编程环境 元计算 计算网格 MPI PVM 代理 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]