Quantifying Distributed System Stability through Simulation: A Case Study of an Agent-Based System for Flow Reconstruction of DDoS Attacks

We investigate the stability properties of a novel agent-based system for the detection of network bandwidth-based distributed denial of service (DDoS) attacks. The proposed system provides a description of the structure of flows which comprise the DDoS attack. In doing so, it facilitates DDoS mitigation at or near attack traffic sources. The constituent agents within the system operate at the inter autonomous system (AS) level, comprising a distributed collection of IP-layer network taps which self-organize in response to attack flows. We formalize the notion of stability for the proposed system, and show how we can use simulation to identify regions of instability within the system´s parameter space. We then modify our system design to circumvent the uncovered singularities, and demonstrate the efficacy and tradeoffs implicit in our redesigned system.