Une caractérisation non gaussienne et à longue mémoire du trafic Internet et de ses anomalies: validation expérimentale et application à la détection d’attaque de DDoS/Non Gaussian Long Memory Model for Internet Traffic: Experimental Validation and Application to DDoS Detection

Abstract

Being now a mainstream communication, Internet is subject to many kinds of anomalies (failures, flash-crowds, attacks). In order to compare the statistics of normal traffic with traffic with anomalies, we collect both regular and anomalous traffic. The traffic is collected on the RENATER network by the METROSEC project and we produce both Denial of Service (DoS) attacks with real attack softwares (TFN2K, TRIN00) aimed at various services (ICMP, SYN, UDP, TCP), and flash-crowd anomalies. We propose a multiresolution, non-Gaussian model with long memory and the corresponding estimators. It models, jointly at all aggregation levels, normal traffic, and also traffic containing anomalies. We show that the model enables to detect the anomalies in the traffic and distinguish between flash-crowd and DoS types of anomaly.