Random early detection (RED) is an effective active queue management (AQM) algorithm controlling network congestion by average queue length. However, RED is over-sensitive to congestion degrees and its parameters. To overcome RED's shortcomings, this paper proposes a dynamic RED algorithm named DRED for enhanced performance. Two improvements are made: (1) DRED replaces the linear packet dropping function with a nonlinear function, which gives differential reactions to different network conditions. (2) Adaptation rule is introduced for the maximum packet dropping probability to stabilize the average queue length. By simulation comparisons, the results verify that DRED outperforms the original RED algorithm in terms of stability and robustness. DRED achieves stable average queue length and predictable queuing delay independent of number of TCP connections. Moreover, DRED solves RED's problems without introducing additional variable or much extra overhead.
|Number of pages||8|
|Journal||Journal of Computational Information Systems|
|Publication status||Published - 1 Dec 2011|
- Active queue management
- Congestion control
- Random early detection