几种主动队列管理拥塞控制算法的比较研究

Comparative Study of Several Active Queue Management Algorithms

主动队列管理是近年来端到端拥塞控制研究的热点,50多种AQM算法已经被提出,但路由器中采用哪种算法没有统一认识。通过仿真实验,对ARED、AVQ、PI和REM 4种主动队列管理拥塞控制算法在相同的仿真网络环境中进行了比较研究。实验研究表明,这4种AQM算法都能使队列稳定在目标值,PI算法使队列最稳定,AVQ算法维持一个较小的队列长度;4种AQM算法都能经过一定时间适应网络变化的要求,PI和AVQ算法有较好的性能,短流对ARED和REM算法有较大影响;4种算法都没有根本解决对UDP流的公平性问题。

Active Queue Management (AQM) has been a recent focus in the research area of end-to -end congestion control. More than 50 AQM algorithms have been proposed. However, there is no agreement yet which algorithm should be used in network routers. In this paper, four important AQM algorithms, i. e. , ARED, AVQ, PI and REM, are compared in the same network environment, by means of network-simulation. The experimental study shows that, while all the studied AQM algorithms can keep queues stable, the queue managed by PI is the most stable one, and AVQ can maintain a shorter queue than ones managed by other algorithms. It is found that, while all the four algorithms can adapt to network conditions, PI and AVQ have better adaptability, and the performance of ARED and REM can be influenced substantially by short-lived TCP flows. Simulation results also show that none of the four algorithms can perfectly solve the unfairness problem for UDP flows.