| 引用本文: | 刘正飞,孙金生,胡斯乔.TCP/AQM系统大时间尺度模型[J].控制理论与应用,2018,35(4):475~484.[点击复制] |
| LIU Zheng-fei,SUN Jin-sheng,HU Si-qiao.A large time scale model of TCP/AQM system[J].Control Theory and Technology,2018,35(4):475~484.[点击复制] |
|
| TCP/AQM系统大时间尺度模型 |
| A large time scale model of TCP/AQM system |
| 摘要点击 2160 全文点击 1027 投稿时间：2017-07-10 修订日期：2017-11-06 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2017.70469 |
| 2018,35(4):475-484 |
| 中文关键词 主动队列管理 多时滞耦合 大时间尺度 广义预测自适应控制 |
| 英文关键词 active queue management multi-delay coupling large time scale generalized predictive adaptive control |
| 基金项目 |
|
| 中文摘要 |
| 网络传输连接的往返时延(round-trip times, RTT)大小各不相同, 因此TCP/AQM系统本质为一多时滞回路
耦合系统. 由于RTT分布范围远大于控制量调节周期, 这给准确评估控制效果带来很大困难. 已有基于控制理论的
主动队列管理(active queue management, AQM)算法多以流体流模型为基础进行设计, 没有充分考虑RTT和采样周
期对系统性能的影响. 对于TCP/AQM系统, 合理的评价方法是对调节过程进行评价, 而非仅评价单个采样周期内的
控制量是否合适. 本文结合数据驱动控制思想和系统自身特征, 统一从路由视角对TCP与AQM之间的交互进行抽
象, 通过时间扩展从更大的时间尺度去评价控制量调节过程, 然后基于此模型设计自适应AQM算法–—大时间尺
度AQM算法(large time scale AQM, LTSAQM). 仿真结果表明, 该算法收敛速度快, 排队时延抖动小, 特别是在长时
滞网络环境下, 性能明显改善. |
| 英文摘要 |
| Since the round-trip times (RTT) of network traffic are diverse, TCP/AQM system is essentially a multi-delay
coupling system. As the distribution of RTT is often far larger than the input regulating period, it becomes difficult to accurately
assess the control effect. Most control-theoretic active queue management (AQM) schemes are designed based on the
fluid-flow model, without considering the effect of large RTTs on system performance. A reasonable model for TCP/AQM
system should focus on the evaluation of input adjustment process, rather than solely on evaluating whether the control input
in a certain period is appropriate or not. Combining the data driven control with plant characteristics, this paper models
the interactions between TCP protocol and AQM from an unified router perspective. Through time expansion, a large time
scale evaluation model of TCP/AQM is obtained. An adaptive AQM scheme named large time scale AQM (LTSAQM) is
then proposed based on the deduced model. Simulation results show the proposed method achieves fast convergence speed,
small queueing delay jitters and prominent performance improvement in long-delay network environments. |
|
|
|
|
|