基于PID神经网络控制的服务器分配方法  

Autonomic Server Provisioning Based On PID Neural Network Control

作　　者：脱立恒[1,2] 倪宏[2] 刘学[2] 

机构地区：[1]中国科学院大学,北京100049 [2]中国科学院声学研究所,北京100190

出　　处：《计算机仿真》2014年第1期302-306,322,共6页Computer Simulation

基　　金：国家科技支撑计划(2011BAH11B04);国家高技术研究发展计划(2011AA01A102);中国科学院战略性先导科技专项子课题(XDA6030500)

摘　　要：研究多层网络应用服务系统的服务器分配问题,由于服务器传输延迟不可控,造成服务系统不稳定。为解决上述问题,提出采用比例积分微分(PID)神经网络控制的服务器分配方法。首先建立多层服务器分配模型,以服务系统处理延迟保证为限制条件,以最小化分配服务器数目为优化目标,将服务器分配问题抽象为最优化问题,并通过最优化方法求解;同时,为了保证95%的服务系统处理延迟在目标延迟以内,通过PID神经网络控制对通过最优化方法得到的服务器分配方案进行调整;并通过设计PID神经辨识系统对被控系统进行辨识,保证系统的稳定性。与现有方法相比,改进方法分配的服务器总数目减小6%,同时,在产生误差抖动和到达率变化时,系统更早收敛,更快恢复稳定。Autonomic server provisioning based on Proportion-Integration-Differentiation （PID） neural network control, PIDN in short, was proposed to deal with server provisioning in service system for multi-layer network appli- cation. We built multi-layer server provisioning model at first. Based on the model, we considered server provisio- ning as optimization problem which minimizes the sum of servers in all layers under the average processing delay guar- antees. We solved the problem using optimization method. In order to ensure that the percent of requests whose pro- cessing delay is less than the target delay in all requests is up to 95%, we adjusted the resuh from optimization meth- ods by adding PID neural network controller. Besides, we designed PID neural network identifier to identify the system controlled to guarantee the stability of the system. The PIDN reduces the total number of servers by 6%. The system converges earlier and restores stability more quickly under jitters and changes on request arrival rate.

关 键 词：服务器分配 神经网络 网络应用 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]