摘要
为进一步提高集群系统服务的性能,在对静态负载均衡和动态负载均衡的优缺点分析的基础上,提出一种基于剩余负载率的动态均衡机制.该均衡机制采用剩余负载率作为负载状态的评价标准;针对服务器各节点性能的不同,提出采用BP神经网络训练节点;并设计了一种基于流表的静态分配策略和基于负载预测的动态分配策略相结合的任务分配策略来实现任务在集群系统各节点间的动态分配,从而降低了服务器各节点之间任务重新调度的次数,提高了集群系统的服务性能.实验结果表明,该均衡机制是可行的、有效的.
A leftover Load Rate - based dynamic balance mechanism is propced after analyzing static balance technology and dynamic balance technology on the purpose of improving the performance of cluster systems. This mechanism adopts leftover Load Rate as appraisable standard; BP neural network is used to train nodes of cluster systems because of different performance of the nodes; and a new distribution method which is based on static balance technology with a flow table and dynamic balance technology based on load - forecasting is proposed to distribute assignments to nodes of cluster systems. This distribution method reduces times of dispatching assignments between nodes, and improves the performance of cluster systems. The final simulation result suggests this balance mechanism is feasible and valid.
出处
《微电子学与计算机》
CSCD
北大核心
2010年第2期141-144,共4页
Microelectronics & Computer
关键词
剩余负载率
负载均衡
BP神经网络
报文
leftover load rate （LLR）
load balance
neural network
message
