基于队列和负载因子的动态参数随机指数标记算法  

A Dynamic Random Exponential Marking Algorithm Based Queue Factor and Load Factor

作　　者：汪浩[1] 田作华[1] 

机构地区：[1]上海交通大学自动化系,上海200240

出　　处：《四川大学学报（工程科学版）》2010年第1期173-178,共6页Journal of Sichuan University (Engineering Science Edition)

基　　金：国家自然科学基金资助项目(60574081)

摘　　要：为了解决随机指数标记算法(REM)队列抖动大,对动态数据流响应慢,以及环境适应性差等问题,分析了算法的控制属性,并提出了一种参数动态调整的随机指数标记算法(DREM)。基于控制理论的分析表明,REM算法具有比例积分(PI)控制属性。通过引入队列因子和负载因子的概念,对队列调整状态进行实时划分,能够有效地判断当前网络的拥塞状况。同时,利用队列和负载因子设计了关键参数的调整率,以协助基于"和式增加积式减少(AIMD)"规则的TCP拥塞控制策略,有效增强了REM算法的控制性能。NS2平台中的仿真实验表明,相对于标准REM算法,DREM提高了队列长度的响应能力,减小了丢包率,增强了主动队列管理算法的适应性和鲁棒性。To solve the problem that random exponential marking （REM） suffered from some drawbacks such as big queue oscillations, sluggish response to dynamic network traffic and poor adaptability to various network conditions, the control property of REM was analyzed, and a dynamic REM （DREM） scheme was p based analysis indicated that REM had the variables, the queue factor and load factor, same property with proportional were introduced to divide the integral control. Additionally, two new regulating procedure of queue length into four cases, which could identify the network status effectively. Moreover, the key parameter of REM was adjusted dynamically by using the queue factor and load factor to assist the ‘additive increase multiplicative decrease （AIMD） ＇ strategy-based TCP congestion control mechanism, and enhance the control performance of REM. Simulation and comparison with original REM in NS2 platform demonstrated that DREM could enhance the responsiveness of queue length, reduce the packet loss ratio, and improve the adaptability and robustness for active queue management.

关 键 词：拥塞控制 主动队列管理 随机指数标记 队列因子 负载因子 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]