Software Defined Networking (SDN) has been preferred over traditional networking due to its dynamic nature in adapting the network structure. This agile nature of SDN imparts non-stationarity in traffic. In this work, we characterize the SDN traffic and study its behavior under dynamic conditions using Augmented Dickey Fuller (ADF) test. Later, we model the SDN under non-stationary conditions using queueing model and solve for average queue length at both controller and switch using Pointwise Stationary Fluid Flow Approximation (PSFFA). The analytical results have been validated through simulations. We develop congestion control algorithm based on (a) Proportional Integral Derivative (PID) control mechanism and (b) Dynamic Random Early Detection (DRED) control mechanism for SDN controller using the fluid flow model. Finally we demonstrate their effectiveness in stabilizing the queue length at the switch and controller under non-stationary conditions. In nut shell our work brings out the importance of the non-stationary behaviour of the traffic in the design and analysis of SDN and its control algorithms.

与传统网络相比，软件定义网络（SDN）具有更佳的适应性，因为它在适应网络结构方面具有动态特性。SDN的这种敏捷性使流量变得不稳定。在这项工作中，我们对SDN流量进行表征，并使用增强Dickey Fuller（ADF）测试研究动态条件下的SDN行为。稍后，我们使用排队模型在非平稳条件下对SDN建模，并使用点向平稳流体流量近似（PSFFA）求解控制器和交换机的平均队列长度。分析结果已通过仿真验证。我们使用流体模型基于（a）比例积分微分（PID）控制机制和（b）SDN控制器的动态随机早期检测（DRED）控制机制开发拥塞控制算法。最后，我们证明了它们在稳定非稳态条件下稳定交换机和控制器队列长度的有效性。简而言之，在SDN及其控制算法的设计和分析中，我们的工作凸显了流量的非平稳行为的重要性。