一种适用于网络处理器的队列管理算法

A Queue Management Algorithm Fit for Network Processors

遵循比例区分服务模型,设计了一种适用于网络处理器的队列管理算法·算法包含两部分,分组入队列时实现丢失率控制的RR-PLR(round-robinbasedproportionallossrate)和分组出队列时实现时延控制的WRR-PAD(WRRbasedproportionalaveragedelay)·算法采用轮循的机制,避免了除法运算和排序操作,具有O(1)的复杂度,而且易于在网络处理器上实现·性能模拟以及实测的结果表明,该算法能有效实现平均分组丢失率和平均排队时延的比例控制,系统的总吞吐率达到了1·125Gbps(每个分组64B,即2·25Mpps)·

Following the model of proportional differentiated services, a queuing management algorithm is presented, which is fit for network processors. This algorithm contains two parts, the one using for loss ratio control while packets come into the queue called RR-PLR （round-robin based proportional loss rate）; and the one using for delay control while packets go out of the queue called WRR-PAD （WRR based proportional average delay） . To avoid division and sort operation, this algorithm uses the round-robin mechanism. Its complexity is O （1） and it＇s very fit for the architecture of network processors. The simulation in NS2 shows that this algorithm can achieve the proportional differentiated service in both average packets loss ratio and average queuing delay. A prototype of the algorithm has also been implemented in the Intel IXP2400 platform. The test result shows that the whole system can reach 1. 125Gbps （64 bytes per packet, i.e. 2.25Mpps） and achieve the proportional differentiated service at the same time.