稳定的随机早期检测方法

Stable Random Early Detection Method

针对随机早期检测(RED:Random Early Detection)对网络时滞、参数设置敏感的问题,提出一种适用于时滞网络的稳定随机早期检测算法(TRED:Time-delay RED)。引入史密斯预估器,以抑制网络时滞对网络性能的影响;采用瞬时队列长度替代平均队列长度作为拥塞指示,加快系统的响应能力;改进RED算法的丢包概率函数为非线性函数,同时自动调整系统参数,以适应网络环境变化。仿真结果表明,TRED算法能成功补偿网络延时,并在不同的时滞环境、不同程度的拥塞环境中保持稳定的队列长度,具有很强的环境适应性,从而保证了良好的网络性能。

In order to solve the problems of RED（Random Early Detection） algorithm,such as sensitivity to time delay and parameter configuration,a stable random early detection algorithm is designed for TRED（Time-delay RED）.TRED introduces Smith predictor to restrict the influence of time delay on network performance.TRED takes instantaneous queue length as congestion indicator instead of average queue length for fast system response.TRED modifies the probability function of RED as a nonlinear function and adapts the system parameter to the time-varying dynamics.By simulation comparisons,the results demonstrate that TRED successfully compensates for the time delay and keeps stable queue length in delay networks with various congestion degrees.TRED exhibits strong adaptability and guarantees satisfactory network performance.