An Efficient Method to Collect Statistics in SDN Using Curvature Based Sampling

Network management involves measurements and control. Software Defined Networking (SDN) provides a higher level of control on network devices. In SDN, the controller can poll the statistics at different granularity with a constant or dynamic polling rate. There is always a trade-off between the accuracy and polling overhead. This paper proposes a new method for dynamic polling, which provides the best trade-off between the accuracy and polling overhead. We use curvature based sampling as the basis for our method and compute the change in the rate of change to decide the polling frequency. We use real traffic traces for the experiments. The experimental results demonstrate that the proposed method provides better accuracy compared to MoMon [1] and CeMon [2]. The proposed method achieves 23% accuracy over MoMon [1] and roughly three to four times better accuracy than CeMon [2].