PRED: Prediction-enabled RED

Abstract

This paper proposes a router congestion control mechanism called PRED (Prediction-enabled RED), a more adaptive and proactive version of RED (Random Early Detection). In essence, PRED predicts its queue length for an early detection of possible congestion alerts in the near future and operates adaptively to the predicted changes in traffic patterns. Typically, PRED does this by first making prediction about average queue length and then using the predicted average queue length to adjust three classic RED parameters maxth, minth, and maxp. The incoming packets after the adjustment are now being dropped with the new probability defined by updated parameters. Due to its adaptability and proactive reaction to network traffic changes, PRED can be considered as a novel solution to dynamically configure RED. Extensive simulation results from NS-2 simulator are presented to verify the performance and characteristics of PRED. © Springer-Verlag 2004.