基于多代理的网格任务调度研究  

Research on Grid Task Scheduling Based on Multi-Agent

作　　者：曾万聃[1] 周绪波[2] 戴勃[1] 游新冬[1] 常桂然[1] 

机构地区：[1]东北大学信息科学与工程学院 [2]清华大学软件学院,北京100081

出　　处：《计算机科学》2005年第6期17-20,共4页Computer Science

基　　金：高等学校博士学科点专项科研基金(20030145017)

摘　　要：随着网格技术的发展,代理技术近年来被用在网格的实现当中。多代理技术用分布式自主结构代替集中式的非自主性结构,具有更强的实时性,特别适合于动态调度,本文提出的网格任务调度系统就采用了多代理的体系结构。由于GLObus已经提供了网格操作系统的大部分功能,本系统构建在Globus之上。在充分利用底层Globus提供的功能并与之结合的基础上,在代理中加入人工智能、知识学习的方法和服务质量实现策略,对网格任务进行灵活智能的调度,实现更好的负载平衡并达到一定的服务质量。采用JATLite来创建该系统,利用该系统可以根据应用背景需求快速定制动态任务调度平台。With the development of Grid technology, agent technology has been utilized in the implementation of Grids. The distributed autonomous infrastructure of multi-agent can achieve better real-time response than the cen- tralized non-autonomous ones, and it is very suitable to dynamic scheduling. Because Globus has provided most of the functions of a Grid operating system, our system is based on Globus. This system not only makes full use of and inte- grates with the functions of Globus, but also utilizes artificial intelligence and machine learning methods and adopts the strategies of QoS, to achieve efficient scheduling of Grid tasks with certain levels of QoS assurance. JATLite is used to build the system. This system provides a platform with which a dynamic task scheduling system satisfying the application requirements can be created quickly and efficiently.

关 键 词：GLOBUS JATLite 任务调度系统 服务质量 多代理技术 网格技术 动态调度 体系结构 操作系统 系统构建 充分利用 人工智能 实现策略 知识学习 自主性 集中式 主结构 分布式 实时性 定制 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术] TP311[自动化与计算机技术—计算机科学与技术]