Performance Evaluation of Queue Management Methods for Congestion Control

摘要

Congestion control is proposed to guarantee the network performance and quality of service. A number of Active Queue Management (AQM) methods, such as Random Early Detection (RED), have been designed for congestion control in the past few years. Most of the AQM methods are obtained and verified with different network configurations, and consequently do not have comparability. Moreover, RED is the only scheme recommended by the Internet Engineering Task Force (IETF) and has good potential in practical application. This paper focuses on RED and proposes a Real-time Dynamic RED (RDRED) algorithm, which modifies the static probability function to an adaptive nonlinear function, and uses instantaneous queue length as the congestion indicator instead of average queue size. Then this paper compares the proposed RDRED scheme with RED and its different valiants, such as Adaptive RED (ARED), Nonlinear RED (NLRED), Dynamic RED (DRED) and Improved RED (IRED). The performance evaluation is done by using the Network-Simulator-2, which provides a convenient and reliable platform for simulating large-scale networks. It is demonstrated that the RDRED algorithm achieves the best performance among all the tested AQM algorithms. It can stabilize the queue length very well while keeping the queuing jitter very small. The simulation results also provide insights into the AQM design, e.g. the importance of congestion indicator, adaptive and nonlinear mechanism being suitable for time-varying TCP dynamics.