HIVE-SIGHT: A Swarm-Intelligence Framework for Anticipating and Mitigating Distributed Denial-of-Service Attacks at Global Scale

Abstract

DDoS attacks have become one of the most enduring and disruptive problems to the current internet infrastructure. The current methods that most people use despite the advancement of machine learning are reactive after an attack is initiated. This paper presents HIVE-SIGHT, a decentralised swarm-intelligence system, which is capable of predicting and preventing massive DDoS attacks before they grow. Based on the collaborative behaviour of biological swarms, the system uses lightweight agents, also known as HIVE Nodes, deployed throughout networks, cloud edges and ISP infrastructures. Such nodes provide each other with bit-sized anomaly indicators, update their behaviour by means of distributed reinforcement learning, and collaboratively create a real-time picture of the threat situation in the world. In the case of testing its performance, we created a multi-ISP dataset of more than 38 million packets of twelve DDoS variants. The experimental analysis reveals that HIVE-SIGHT is capable of predicting attack peaks with a maximum of 270 seconds and has a prediction accuracy of 92.7 percent and mitigation latency is almost half that of traditional AI-based defences. Such results indicate that swarm-intelligence principles can present a future solution in the form of resilient, self-organising ecosystems of cyber defence systems with the capability to scale to internet scale.