一种低管理开销的网格计算模型  

Low management cost grid computation model

作　　者：何凯[1] 杨学刚[1] 周庶恺[1] 杨愚鲁[1] 

机构地区：[1]南开大学信息技术科学学院计算机科学与技术系,天津300071

出　　处：《计算机工程与应用》2008年第8期123-127,共5页Computer Engineering and Applications

基　　金：国家自然科学基金( the National Natural Science Foundation of China under Grant No.60473088);天津市科技发展计划资助项目( No.05YFGZGX2390)

摘　　要：针对利用Internet上大量空闲计算资源来解决大规模分布式计算问题这一需求,提出了一种低管理开销的网格计算模型。在该模型中,不存在任何节点来管理动态变化的资源,而与之相适应的信息机制、任务调度算法和有限任务复制算法在没有管理节点存在的情况下,以较低的开销使系统在动态的环境中达到自然的协调,实现大规模的分布计算。开发的仿真软件验证了该模型的有效性,并对相关结果进行了初步的性能分析;仿真结果表明,该模型在动态的环境中负载分布合理,资源的计算能力能得到充分利用,为高效地完成参数扫描、蒙特卡罗模拟等大规模易并行计算提供了一个可行的方法。This paper proposes a low management cost computing grid model,aims at harvesting the idle time of Internet connected computers to run very large distributed applications.In this grid model,no specific node is designated to manage the dynamically changing resources.In the absence of the management node,the corresponding information mechanism,task schedule algorithm and limited task replication algorithm naturally evolve the system into equilibrium to perform large scale distributed computation with low cost.A simulation software package has been developed to verify this model,and to assess its performance.The results show that the model reasonably distributes loads in a dynamic environment,and fully utilizes the computational capacity of the idling resources,it is a high-efficiency resolution for large scale,embarrassingly parallel applications such as parameter sweep and Monte-Carlo simulations.

关 键 词：计算网格 低管理开销 任务调度 信息机制 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]