Ola Marwan Assim, Qutaiba I. Ali, Zahraa T. Al Mokhtar, Nawal Y. Abdullah
This paper presents an approach that incorporates classical Artificial Intelligence techniques inside the software Defined Networking (SDN) for traffic classification and makes the controller more responsive besides adding intelligence to the network. Three of the most popular machine learning classifiers frequently used by researchers are evaluated: Naïve Bayes, Support Vector Machine, and Nearest Centroid. These are either embedded or attached to three leading SDN platforms (Ryu, ONOS, and OpenDaylight). An elaborate experimental methodology using Mininet was developed to test several configuration parameters and policies both in a real environment as well as under synthetic conditions. Results show that out of all flows classified with over 94% accuracy by Naïve Bayes classifier keeping decision latency at only 8.2 ms on the controller, meanwhile traditional SDN setup with AI has a decision latency of 15 .7ms. The Open Network Operating System (ONOS) showed a maximum sustained bandwidth of 85 Mbps with AI help while reducing the resource usage by as much as 30% on externalized feature extraction. This result, therefore, clearly proves that lightweight AI has emphasized impacts due to proper parameterization without incurring huge additional computational costs. The best project’s efforts are in proposing methods for the selection of controllers and tuning parameters together with hybridization between AI and SDN network strategies suitable either for real-time or large-scale networks