On the use of information theory metrics for detecting DDoS attacks and flash events : an empirical analysis, comparison, and future directions

Abstract EN

A distributed denial of service (DDoS) attack is one of the lethal threats that can cripple down the computing and communication resources of a web server hosting internet-based services and applications.

it has motivated the researchers over the years to find diversified and robust solutions to combat against DDoS attacks and characterization of flash events (a sudden surge in the legitimate traffic) from HR-DDoS (High-Rate DDoS) attacks.

In recent times, the volume of legitimate traffic has also magnified manifolds.

It results in behavioral similarities of attack traffic and legitimate traffic that make it very difficult and crucial to differentiate between the two.

predominantly, netflow-based techniques are in use for detecting and differentiating legitimate and attack traffic flows.

over the last decade, fellow researchers have extensively used distinct information theory metrics for Netflow-based DDoS defense solutions.

however, a comprehensive analysis and comparison of these diversified information theory metrics used for particularly DDoS attack detection are needed for a better understanding of the defense systems based on information theory.

this paper elucidates the efficacy and effectiveness of information theory-based various entropy and divergence measures in the field of DDoS attack detection.

as part of the work, a generalized netflow-based methodology has been proposed.

the proposed detection methodology has been validated using the traffic traces of various real benchmarked datasets on a set of detection system evaluation metrics such as detection rate (Recall), Precision, F-Measure, FPR, Classification rate, and Receiver-Operating Characteristics (ROC) curves.

it has concluded that generalized divergence-based information theory metrics produce more accuracy in detecting different types of attack flows in contrast to entropy-based information theory metrics.