Real-Time Flow Identification Based on Neural Network and OpenFlow Over SDN
Xiaosong Xie, Jincheng Wu · 2018
This paper introduces a novel approach to flow identification by applying a backpropagation Neural Network over SDN. Thanks to the advantages of the global abstract of SDN controller, we get a convenient way to collect data from data-path and implement a real-time predictor of traffic types. Some flow features are selected from OpenFlow counters for each flow. In addition, we get some other flow features by intermittently duplicating flows and leading them to the controller. The data from OpenFlow counters and duplicated flows are integrated as input vector for Neural Network. Especially, we take into consideration packet length distribution and connection count with the same source and destination IP, which improves the accuracy and efficiency of video flow and P2P flow identification.