QRED: A Q-Learning-based Active Queue Management Scheme

Abstract

The Active Queue Management (AQM) algorithm is one of most important research fields in network congestion control. To adjust the maximum dropping probability (maxp) according to the network situation the maxp calculation based on the RED algorithm is improved using the Q-learning algorithm, and a new algorithm, known as QRED (Q-learning RED), is proposed. The self-adaptive adjustment for the maxp is achieved using the QRED algorithm and the queue length stability in a dynamic network environment is realized. In addition, the QRED algorithm not only avoids the sensitivity of the RED algorithm parameters, but also adapts the packet loss rate according to the specific network service type. Results based on the NS2 simulation show that the QRED algorithm has better stability in complex network environments, and hence, are superior to the RED active queue management algorithm.