Multi-feature Enterprise Traffic Characterization in OpenFlow-based Software Defined Networks

Software defined networking (SDN) decouples the data forwarding plane from control logic, enabling real-time traffic engineering, difficult to realize in conventional networking. The prominent southbound OpenFlow protocol used in SDN environments provides an intuitive means for traffic monitoring. Limited statistical information based on generic OpenFlow flow measurements however, fails to represent the correlation among several network parameters of administrative interest, used to construct optimal policies in modern enterprise SDN. The present study investigates the multi-dimensional nature of network traffic in enterprise SDN to realize increased control granularity and provide an insight into network behaviour. OpenFlow based flow statistics are collected for this purpose from a realistic enterprise SDN over a two-week time frame. The flow measurements are independently subjected to unsupervised cluster analysis to derive unique traffic classes based on application trends, aggregate daily traffic volume, flow size, and spatial as well as temporal traffic distribution. Contrary to basic flow-level information, traffic distribution across multiple features provides a comprehensive understanding into the network workload, recommending the utilization of derived traffic classes in several areas of SDN improvement including anomaly detection, energy utilization, load balancing and user-centric policy construction.