Simulated annealing Q-learning algorithm for ABR traffic control of ATM networks

One of the fundamental issues in asynchronous transfer mode (ATM) networks is the congestion problem of information flow. Due to the complexity and variability of ATM, it is difficult to accurately describe the characteristics of source traffic. This paper presents a traffic controller to solving the congestion problem by using Q-learning conjunction with simulated annealing. In stead of relying on the mathematical model for source traffic, the controller is designed to learn an optimal policy by directly interacting with the unknown environment. The simulated annealing is a powerful way to solve hard combinatorial optimization problems, which is used to adjust the balance between exploration and exploitation in learning process. The proposed controller forces the queue size in multiplexer buffer to the desired value by adjusting the source transmission rate of the available bit rate (ABR) service. Simulation results show that the proposed method can promote the performance of the networks and avoid the occurrence of congestion effectively.