A deductive approach for the sensitivity analysis of software defined network parameters

With the exponential growth in the number of internet-enabled devices, large scale security threats such as distributed denial of service (DDoS) attacks significantly affect software defined networks (SDNs). This necessitates efficient detection and mitigation solutions. Monitoring of SDN activities (typically identified using metrics such as throughput, jitter and response time) to ascertain deviations from profiles of normality (previously learned from benign traffic) is a key approach in detecting attacks on SDNs. In this paper, local sensitivity analysis (LSA) is implemented to identify the key network metrics that mainly influence the prediction of whether an SDN is under attack or secure. Using throughput, jitter and response time as the network impact metrics and a mathematical cost function based on min-max feature scaling to associate SDN scenarios with their respective SDN impact metrics, an artificial neural network (ANN)-based prediction model is built. The sensitivity of throughput, jitter and response time is then evaluated using the deviations of newly predicted target values of the ANN model from the actual target values when an additive white Gaussian noise (AWGN) is added to the respective impact metrics. The results of this study show that throughput, jitter and response time are all statistically sensitive to a DDoS flooding attack of the SDN. Also, Jitter was found to be the most sensitive network metric to a DDoS flooding attack of the SDN. © 2020 Elsevier B.V.