Integrated framework for controller placement and attack-aware strategy in SDN-enabled wide area network

Abstract

Software Defined Networks (SDNs) are a promising networking paradigm for managing and controlling large-scale or complex networks, such as wide area networks (WAN). However, SDN also presents new challenges including controller placement problems (CPP) and the vulnerability to cyberattacks. CPP finds the optimal locations and numbers of controllers in the network, considering various performance and reliability systems of measurement. Several controller placement solutions and attack-aware strategies have been proposed to achieve network performance and protect the SDN plans from malicious attacks respectively. Nevertheless, these solutions are disjointed and there is no robust model that integrates them to effectively identify anomaly-based attacks and make the routing and detection process more intelligent and autonomous for traffic analysis. This study proposed and implemented an integrated framework for addressing both CPP objectives and attack-aware strategy in SDN-based WAN. The study carried out a comprehensive literature review to identify different approaches and empirically analyse them to select the optimum models and apply the design science research methodology to realize the framework. The framework consists of two main components: a machine learning (ML) based clustering model, enhanced DBSCAN model for the CPP, which minimizes both worst-case and average latency and maximizes the resilience of the control paths; and an attack-aware strategy for the SDN control plane, which leverages supervised ML-based approach, the k-nearest neighbour (kNN) to design an intrusion detection system (IDS). The integrated framework is evaluated via extensive simulations and experiments using the LAN dataset and various performance evaluation metrics including network latency, reliability, detection accuracy and precision.The results obtained show that the integrated framework can improve the performance and security of SDN-enabled WANs. The enhanced DBSCAN model achieved the optimum number and location of controllers in a given WAN while the kNN-based IDS achieved a high detection accuracy of 100% compared to other models such as Random Forest and Naïve Bayes in the detection of network anomalous. About 8 anomaly attacks were identified in the network while considering the 1.096 threshold for traffic monitoring and identification based on the 95% percentile. Based on these results, it shows that a practical solution to CPP and other related problems of the SDN network should be addressed collectively rather than in isolation since it may depend heavily on solutions to other problems such as security, load balancing, resource requests or scheduling problems. Moreover, a hybrid approach of both supervised and unsupervised ML techniques is effective in developing a robust framework that effectively integrates CPP with IDS to enhance network performance and quality of service while simultaneously defending the network against all forms of malicious network attack