A New Approach to Estimate RED Parameters Using Function Regression

Random Early Drop (RED) is widely applied in Mobile Ad-hoc NETworks (MANETs) for congestion control. It randomly drops packets to prevent congestion from occurring, while keeping proper queue size of Route Request (RREQ) packets at the same time. Unfortunately, in the case of severe congestion and large amounts of RREQ packets in the queue, RED algorithm with tune parameters is not able to improve the congestion status properly. This paper proposes a Dynamic RED (DRED) mechanism based on the fitting curve of packet delivery ratio and packet sending rate, which can be readily achieved in practice. DRED presupposes threshold value for initiating and dynamically changes it according to the function regression. Compared to First In First Out (FIFO) and RED with tune parameters mechanisms, simulation results show that DRED has better performance in terms of average end-to-end delay, Hello packet overhead and packet delivery ratio, while avoiding obvious increase of routing discovery frequency.