一种改进的显式拥塞指示算法

An Improved Explicit Congestion Notification Algorithm

RFC 3168中的显式拥塞指示(ECN)算法建议,当网关缓存的平均队列长度超过队列门限上界时丢弃分组,该策略会导致过低的网络吞吐量.为提高系统性能,文中提出一种名为长队列随机丢弃(LQRD)的改进算法,该算法在网关缓存中的平均队列长度超过队列上限时按一丢弃概率函数随机地丢弃分组以限制队列长度继续增加,其中丢弃概率函数综合考虑了平均队列、网关的缓存大小、链路容量和系统平均延迟等因素的影响.仿真结果表明,文中算法在网关缓存保持较长队列时,能保证高吞吐量、低延迟和连通的鲁棒性.最后,通过非线性差分方程分析了该算法的解析模型.

As the algorithm of explicit congestion notification （ECN） in RFC 3168 recommends the gateway to drop packets when the average queue length in the gateway buffer exceeds the upper queue threshold, the network throughput may be too low. In order to improve the system performance, this paper proposes an algorithm named Long-Queue Random Drop （LQRD）. In the proposed algorithm, the gateway drops packets according to a drop probability function when the average queue length in the gateway buffer exceeds the upper queue threshold, thus limiting the continuous increase in queue length. The drop probability function synthetically considers the average queue length, the buffer size, the link capacity and the average queue delay. Simulated results show that the proposed algorithm ensures high throughput, short delay and robust connectivity for long queue length in the gateway buffer. The analytical model of the proposed algorithm is finally analyzed by using the nonlinearly differential equation.