Active Queue Management Design and Development based on the Correlation of Internet Traffic

A congestion control mechanism is proposed, correlation based random early detection, or CRED that exploits the fact that Internet traffic is often long range dependent and self similar. It thus has a past history that can be used to estimate the mean queue length in the not too distant future. This enables CRED to detect the congestion degree status, either increasing or decreasing, and react accordingly. It is demonstrated that this can result in substantial performance gain when compared with existing versions of RED. CRED shows lower delay with the same loss rate and throughput and the mean queue length profile is more stable with light or heavier congestion.