Adaptive REM: random exponential marking with improved robustness

Introduction: As a representative active queue management (AQM) scheme, random exponential marking (REM) [1] faces the challenge of dynamically tuning its parameters to satisfy the stability requirement under various network conditions adaptive to network conditions [2, 3]. In [4], Deng et al. proposed an adaptive control mechanism for the proportional and integral (PI) controller based AQM (PI-AQM) developed in [3]. The mechanism improves system stability and performance under changing network conditions by applying an update function to the packet dropping probability. Unfortunately, although REM also belongs to the family of PI controllers [4], due to the fact that the proportional control gain KP of REM is a decreasing function of the equilibrium point of marking probability, the mechanism proposed in [4] is not suitable to enable REM to adapt its control parameters to the changing of network conditions. To improve the robustness of REM, in this Letter we propose to apply an update function to the dynamics of the price in REM. The enhanced REM, termed as ‘adaptive REM’, can adapt its control parameters to various network conditions.