GSM:一种多服务器系统的动态负载均衡调度模型  

GSM:a Dynamic Load Balance Scheduling Model of Multi-Server Systems

作　　者：陈进才[1] 张江陵[1] 冯丹[1] 

机构地区：[1]华中科技大学外存储系统国家专业实验室,湖北武汉430074

出　　处：《小型微型计算机系统》2004年第7期1260-1263,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金 (60 2 73 0 74)资助;霍英东教育基金项目(序号 90 10 68)资助

摘　　要：设多服务器系统由 n个服务器结点和 1个中心任务调度结点组成 ,所有服务器具有同样的处理能力 ,任务的到达是一速率参数为 nλ的泊松流 ,任务的服务器处理时间服从参数为 1/μ的指数分布 (λ<μ) ,一次负载信息收集时间和任务调度时间在理想情况下可忽略不计 .基于这些假设 ,提出了多服务器系统的广义超级市场调度模型 (GeneralizedSuperm arket Model,GSM) ,详细分析并证明了该调度模型的解特性 .结论表明 :对于大规模多服务器系统 ,广义超级市场调度模型与随机选择服务调度策略相比 ,可指数级地提高平均任务消耗时间性能 ;与文献 [1,2 ,3]的结果相比 。Suppose: ① a multi server system is composed of n service nodes and 1 central load balance scheduling node; ② all of the servers have the same processing capability; ③ the arrivals of tasks are a Poisson stream of rate nλ; ④ the service time for a task on a server is exponentially distributed with mean 1/μ, here λ<μ; ⑤ both load information gathering time and task scheduling time can be omitted under ideal conditions. Based on these assumptions, a Generalized Supermarket Model (GSM) is presented, and the solution properties of the model are analyzed and proved in details. The result shows that: by use of the Generalized Supermarket Model scheduling strategy for large scale of multi server systems, comparing with that of random service node choice strategy, the mean loss time for a task on a server can be exponentially decreased; and comparing with the results of literatures , it can reach the minimum of them.

关 键 词：多服务器系统 负载 动态负载均衡 调度 模型 

分 类 号：TP393.0[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]