Lightweight deep learning for real-time DDoS detection in SDN using programmable data planes

Distributed Denial of Service (DDoS) attacks are a very continuous and ever-increasing threat against the Software-Defined Networking (SDN) environments as the centralized control plane serves as the most essential vulnerability point in these domains. This research provides a lightweight deep learning mechanism for the realtime detection of DDoS attacks, where DDoS detection is carried out on a programmable data plane, offloading from the SDN controller. A compact Convolutional Neural Network (CNN) model is deployed at the P4-enabled switches such that it could enable high-speed, low-latency detection that could be suitable for resource-scarce devices. The experiments were performed on standard datasets that showed detection accuracy of more than 99% with significantly reduced latencies and resource consumption in detection, proving beyond doubt its efficiency when compared with existing state-of-the-art mechanisms. This paper provides an all-inclusive discussion on the architecture, methodology, results, and future implications for SDN security deployments.