Stable Tuning for Random Early Detection Algorithm

Yang Lin,Zhu Jin,Xie Wanqing,Tan Xiaobin
DOI: https://doi.org/10.1109/chicc.2014.6895874
2014-01-01
Abstract:Although Adaptive RED (ARED) algorithms, which can adjust RED's parameters adaptively are broadly researched to deal with dynamic network scenarios, including bursty traffic and varying link state, unstability is still a constraint against the improvement of efficiency. In this paper, TCP/RED system is considered as a feedback system where equilibrium exists. Through analyzing the proposed system, the adverse effects of busty traffic and varying link state are studied respectively. We also point out that, exist ARED algorithms are low-efficiency because there is equilibrium drifting during adjustment in these algorithms. Therefore, we propose a stable tuning manner for RED parameters which can keep the equilibrium point fixed faced with bursty traffic. In our proposed algorithm, the target queue size is set along with the measured equilibrium point in real time, and the adjustment method for RED parameters is by tuning the slope of the packet dropping probability. This method can keep the equilibrium point in avoidance of equilibrium point drifting under various traffic conditions. Numerical results in NS3 prove our algorithm is more stable and efficient.
What problem does this paper attempt to address?