一种顽健的自校正主动队列管理机制

现有基于控制理论的主动队列管理机制（AQM）大多数是根据简化的线性被控对象模型设计的,或者根据特定的网络条件设置算法的参数,当网络条件大范围变化时算法的性能难以保证.为了解决这些问题,提出了一种自校正的主动队列管理机制STR,通过在线估计TCP/AQM闭环系统被控对象模型的参数,并相应地调节报文丢弃概率,使路由器的缓冲区队列长度与期望值之间的方差最小.通过仿真实验验证了当网络条件大范围变化时算法的队列长度、链路利用率、报文丢弃率等性能,实验结果表明该算法具有良好的顽健性.

Robust self-tuning active queue management mechanism

Most of the active queue management (AQM) mechanisms proposed previously were designed based on the simplified linear model of the controlled object. Moreover, their parameters were usually set according to some specific network conditions. So these algorithms had difficulty in obtaining desirable performance once the network conditions changed widely. To address these issues, a new self-tuning AQM algorithm called STR was proposed. STR had the ability of keeping minimum variance between the instantaneous queue length of the router and the desired value by estimating the parameters of the model of controlled object online and adjusting the packet drop probability accordingly. At last, the performance of STR was evaluated through extensive simulations. The results show that STR is robust against the great changes of the network parameters as well as the traffic load.