Adaptive Active Queue Management Algorithm

After decades of rapid development, the computer network has been used widely. Today, the fast development has created a system of such a magnitude, also to the current computer network has brought a lot of defects which can not be avoided. Rapid growth of network capacity can not meet even more rapid growth of user demand, many bottlenecks arise, therefore, one of which is the network congestion. Phenomenon usually is:the actual effective throughput decreases, the network delay grows, etc, even network crash when serious.To solve this problem, there have been many historical network congestion control algorithms, the focus of current research is the active queue management (AQM) algorithm.This paper describes the causes of network congestion and the research history and present situation, then describes some of the classical congestion control algorithms. Seeing the great variability and strong non-linear of network, I design two algorithms with certain adaptive ability.1) The fuzzy algorithm has good dynamic characteristics, the overshoot is low, system response time is short, but the steady-state error is relatively large. PI algorithm work better in steady-state, steady-state error is small, but overshoot is large, response time is long. The first algorithm is based on the error of real-time queue length and expected queue length, to achieve a gentle switch between the fuzzy algorithm and the PI algorithm, in order to take full advantages of the characteristics of this two algorithms. From the simulation results created by NS-2, compared with the traditional PI algorithm, the dynamic performance and steady-state performance of this new algorithm are improved to some extent.2) REM algorithm uses the queue length deviation and the deviation of input and output rate to calculate a price of link, according to this price to calculate the probability for droping packet. From the view in control theory, the rate is the differential of the queue length. So compared to the congestion control algorithms just using the queue length, REM algorithm has a better ability to forecast a network congestion. But the simulation shows that REM algorithm generates a larger steady-state error and great fluctuations of queue length. To reduce the steady-state error of REM algorithm, I added the proportion algorithm into REM algorithm. When the system enters steady state, the proportion algorithm plays a major role. By the NS-2 simulation results, compared to REM algorithm, the new algorithm performs better in steady-state and dynamic-state.