Abstract: Random Early Detection algorithm in traditional field of congestion avoidance is based on Poisson model，thus it is not adapted to the traffic of self-similar.In this paper，based on self-similar traffic，we propose a RED algorithm，BIRED（Based on Interval Random Early Detection）.According to the self-similarity，by controlling the average queue length，the BIRED can help the network avoiding congestion.Comparing the BIRED and RED about the dynamic performance，the BIRED is superior in robustly.As evident from the simulation，the BIRED performs more better，adapt to the self-similar traffic，controlling the average queue length，illustrating improved transient performance.

Key words: self-similar, active queue management, Random Early Detection（RED）, congestion avoidance

摘要： 传统拥塞避免机制所采用的随机早期检测（Random Early Detection，RED）算法是建立在网络流量按Poisson分布的基础上，因此不适应具有自相似业务流量特征的Internet网。针对自相似业务流量特点，对RED算法进行了改进，提出了一种基于时间间隔的随机早期检测（Based Interval Random Early Detection，BIRED）算法。BIRED算法能够在自相似业务流量特征下，通过控制平均队列长度，避免网络进入拥塞状态。通过仿真比较了BIRED和RED的动态性能，显示了BIRED具有更好的鲁棒性，证明了BIRED比RED、DropTail更能适应自相似业务流量，减小系统负担，有效稳定平均队列长度，改善系统的暂态特性。

关键词: 自相似, 主动队列管理, RED, 拥塞避免