基于改进模型的网络拥塞控制策略

Network Congestion Control Strategy based on the Improved Model

针对基于流体流理论提出的网络模型在推导过程中的一些近似使得该模型对网络行为描述的不精确问题,提出了网络流量的改进模型,并且基于该模型把一种新的类Proportional IntegralDifferential(PID)设计方法用于主动队列管理(AQM)控制器的设计,利用带约束的数值优化方法寻找控制器参数.理论分析和仿真实验表明,该控制算法的综合性能优于已有的Random Early Detection(RED)、Proportional Integral(PI)等算法.表现为平均队列长度更趋于期望值;调节时间更短;队列长度的抖动更小;抗突发业务流干扰能力更强;自适应能力更强.

Because network model based on fluid flow theory can properly describe actions of transfer control protocol （TCP） transportation flow, it is widely utilized by researchers. With this model, Active Queue Management （AQM） algorithms such as Proportional Integral （PI）, Proportional Integral Differential （PID） and improved PID are obtained. These algorithms improve congestion performances of network. However, when they are confronted with large time delay and small expectation queue length, performances will decline. Aimed at improving imprecision of description based on network model for network actions in some circumstances, we present an improved model of network fluid flow in this paper, which applies PID and PID-like controller design methods to AQM controller design to build a novel PID congestion control algorithm. A numerical optimization algorithm is used to search the controller parameters. Taking parabola domain of 4σ＋ω^2 ＋e ≤ 0 as D stable domain, according to control theory, the system will achieve fairly good dynamic and steady performances when diagnostic roots lie to the left of parabola. Theoretical analysis and simulation experiments show that, compared with algorithms such as RED and PI, average queue length of the novel algorithm will be of more approximate expectation, dithering will be smaller, link usage rate will be higher and regulation time will be shorter.