A Rectified Queue-length-based Approach to Active Queue Management for Enhancing Network Performance

Abstract

A lthough, the simple mathematical composition and algorithmic execution of Random Early
Detection (RED) algorithm has been commended by scholars and experts in the research community
around the globe; however, the enforced linear increase probabilistic drop ping central to its operation is continually being revisited and improved, intensifying the proposal of a new spectrum of contributions to finding an optimal dropping function(s) in the recent times. In this paper, a new reworking to RED algorithm called R ectified RED (Rec RED) which splits the threshold regions of RED into three zones is presented. As such, Rec RED, which commingles three smoothly joined dropping functions, namely, cubic, linear, and exponential as a useful replacement for the conventional linear growth function implemented in RED is presented. Rec RED optimizes dropping probability by dropping very few packets at light network loads and scales up its aggressiveness during heavy traffic. Simulation experiments performed in ns 3 software sho ws that Rec RED effectively mitigates queue oscillation and offers an increased network performance. Felicitously, upgrading from RED to Rec RED requires less complication in implementation efforts as simple fine tuning of the dropping profile section of R ED implementation is necessary.

Keywords AQM; Congestion control ; Packet routers ; Rec RED ; RED algorithm