Abstract: As the calculation of the packet loss probability is lack of accuracy without considering the self-similarity of the network flow, a new packet discard model based on the average changing rate of the packet enqueue rate and the ideal queue length is brought out, and it gives the corresponding algorithms—DRED algorithms. By introducing the state of the network flow into the calculation of the packet loss probability, it overcomes the drawback that the RED directly discards all the packets that will enqueue when the average length of queue between the maximum threshold and the maximum length, and the packet discard probability varies with the changes of the network traffic states. The?simulation results show that to adopt the DRED algorithms can make the calculation of the packet lose rate more accurate, the throughput of the network increases, the packet loss rate reduces and the network performance is improved, although the end-to-end delay increases slightly, but the delay jitter is relative smaller, compared with the RED algorithms.

Key words: average changing rate, ideal length, queue, packet discard model, Developed Random Early Detection（DRED）

摘要： 针对RED队列丢包概率模型在计算丢包概率时精确性不足且未考虑网络流量的自相似性问题，提出了基于数据包入队速率平均变化率和队列空闲长度的队列丢包概率模型（DRED），给出了相应的实现算法。DRED将网络流量状态引入到丢包概率的计算过程中，丢包概率随着网络流量状态的变化而变化，克服了RED队列丢包概率模型在平均队列长度大于队列最大阈值小于队列最大长度时直接将到达的数据包全部丢弃的弊端。实验结果表明，与RED相比，DRED丢包概率的计算更加精确，丢包率有所降低，吞吐量相对提高，端到端时延虽稍有增大，但时延抖动较小，网络的整体性能有一定提高。

关键词: 平均变化率, 空闲长度, 队列, 丢包概率模型, 改正的数据包随机早期丢失算法（DRED）