Abstract:

Based on the theory of model predictive control, a novel Active Queue Management (AQM) algorithm is proposed for typical Internet systems with large delay, called MPAQM. First, the predicted output is defined by considering the causality of time-delay system. Then, according to the predictive model and state estimator, the future dynamic queue length in data buffer is predicted. Finally, the control requirement of congestion is converted to optimal control objective function, and drop probability is obtained by solving the optimal problem. The performances of MPAQM are evaluated through a series of simulations in simulator NS2 with single and multiple-bottleneck network topologies. Simulation results show that the proposed MPAQM algorithm outperforms RED and PI algorithms in terms of stability, disturbance rejection and robustness.

Key words: automatic control technology, control theory, network congestion control, active queue management, model predictive control