自适应的PIP主动队列管理机制  

Design of an Adaptive PIP Algorithm

作　　者：刘明[1] 窦文华[1] 张鹤颖[1] 

机构地区：[1]国防科学技术大学计算机学院,长沙410073

出　　处：《计算机研究与发展》2007年第2期201-207,共7页Journal of Computer Research and Development

基　　金：国家自然科学基金项目(90104001);国家"九七三"重点基础研究发展规划基金项目(2003CB314002)

摘　　要：近年来AQM的研究者提出了多种主动队列管理机制,包括RED,PI,REM,AVQ,PD,SMVS,PIP等,它们之间的主要区别在于丢弃概率的计算方法不同,其中基于反馈校正的PIP是综合性能更为突出的一种算法,但是遗憾的是其参数不能实现自动配置.结合单神经元自适应PID控制器,为PIP算法建立了自适应的模型,提出一种参数自适应的PIP算法.通过NS2仿真实验,验证了该算法能提高链路利用率和降低报文丢失率,有效缓解了根据特定网络条件配置算法参数的问题.结合PI,REM,AVQ,PD等AQM算法,讨论了该自适应模型在其他AQM机制中的推广.Active queue management （AQM） is an effective method to improve the performance of end-toend congestion control. Several AQM schemes have been proposed to provide low delay and low loss service in best-effort networks in recent studies, such as RED, PI, REM, AVQ, PD, SMVS and PIP. Among them, PIP is the fusion of PI controller and position feedback compensation and shows better performance under most network conditions, but its parameters can not change with the environments. Based on adaptive single-neuron PID controller, an adaptive PIP AQM scheme is developed using square error of queue length as performance criteria to consolidate the advantages of single neuron and PIP controller. Verified by using NS-2 simulations under a variety of network and traffic situations, the adaptive PIP can achieve faster convergence speed and smaller queue oscillation than PIP, PI, ARED and SPI （selfconfiguring PI, which is an improved algorithm of PI）. In addition, the adaptive scheme can also be used in PI, REM, AVQ, and PD schemes and offers the possibility of optimizing these AQM schemes.

关 键 词：拥塞控制 主动队列管理 单神经元 鲁棒性 控制理论 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]