Toward Joining DDoS Mitigation and Image Segmentation

Zusammenfassung

Volumetric Distributed Denial of Service attacks constitute a significant threat in today’s Internet, as attackers can deny legitimate users access to online services. To mitigate these attacks, i.e., filtering as much attack traffic as possible while preserving legitimate traffic, we propose a novel mitigation approach that copes with large traffic volumes by aggregating ingress network traffic as two-dimensional images. The images serve as input for image segmentation, determining precise IP-based filter rules. Leveraging image segmentation enables powerful machine learning models to achieve high filtering precision. We show the approach’s feasibility by evaluating filtering precision with authentic, real-world traces from CAIDA and MAWI.