基于延迟探测机制的网关队列管理算法

提出一种基于延迟探测机制的算法，该算法通过探测分组在瓶颈网关队列中的延迟时间来估计网络的拥塞状况，设置一个探测门限。当分组的排队延迟超过门限时，采用显式拥塞指示(explicit congestion notification, ECN)机制标记分组以向传输控制协议(transport control protocol, TCP) 源端通知拥塞。为了使该算法在网关中更易实现且具有自适应性，算法在每个周期中都根据不同TCP流对网关资源的占用情况来评估其权重，然后选取权重最大的若干TCP流的测量结果来更新拥塞探测门限，使网关对于突发流量在达到高吞吐量、高链路利用率和稳定的平均队列长度的同时，能更公平地分配资源。通过ns 2下的仿真，证明算法能达到预期的效果。

Gateway queue management algorithm based on delay detection mechanism

An algorithm based on the delay detection mechanism is proposed， which estimates the state of congestion by monitoring the packet queueing delay in the bottleneck gateway and sets a detection threshold. Once the queueing delay of a packet exceeds the threshold, the packet will be marked according to the mechanism of explicit congestion notification (ECN) to notify the transport control protocol (TCP) source. In order to be self-adaptive and easier to perform in the gateway, the algorithm evaluates the weight of different TCP flows according to their occupancy of the gateway resources in every cycle and chooses the measurements of several TCP flows whose weights are the heaviest to update the congestion detection threshold. Through the algorithm, the gateway is not only capable of achieving high throughput, high link utilization and stable mean queue length, but also able to allocate resources more fairly for bursty traffic. The simulation result on network simulator version 2(ns-2) shows that the algorithm can achieve desired performance.