鲁棒的主动队列管理新算法

A Novel Robust Algorithm for Active Queue Management

作为对终端系统上拥塞控制的一种补充 ,中间节点上的主动队列管理 (AQM)策略在保证较高吞吐量的基础上有效地控制队列长度 ,从而实现了控制端到端的时延 ,保证 Qo S的目的 .随机早期检测 (RED)及其派生算法虽可以实现 AQM的技术目标 ,但算法的设计要么是依赖于直觉的 ,要么将网络视为时不变定常系统 ,使得最终形成的算法在稳定性和鲁棒性方面存在不少问题 .在该文的研究中 ,作者将 TCP的流量控制过程等效为二阶线性时变系统 ,采用滑模变结构这种不敏感控制对象参数和负载扰动的控制器设计方法 ,设计出一种新的 AQM算法 .仿真试验表明它的稳态和瞬态性能都优于与它具有相同实现复杂度的 RED和 PI算法 ,并且在负载扰动和参数变化时具有很强的鲁棒性 .

Acting on the intermediate nodes, Active Queue Management (AQM) either keeps the high throughput for routers or effectively controls the queue length, so as to control end-to-end delay for QoS guarantees. With the ability of AQM, Random Early Detection (RED) and its variants have some inevitable flaws, such as unstable and weak, because their designs are excessively dependent on the intuition, which is always unreliable. In this study, authors model the TCP flow process as the two order linear varying-time system, and design a novel AQM algorithm based on the sliding mode variable structure(SMVS) theory, which is insensitive to the parameters of the plant and load disturbance. The simulation results show that its transient and steady performance are superior to ones of RED and PI algorithms. Moreover, SMVS algorithm is very robust against load disturbance and parameter variance.