基于多代理的网格任务调度研究

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。

Research on Grid Task Scheduling Based on Multi-Agent

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.