Fault detection through multi-fractal nature of traffic

Significant progress has been developed previously in modeling network traffic with fractals. These developments have given rise to a new insight and physical understanding of the effects of scaling properties in measured network traffic. Among them, multi-fractal models fit measured data more naturally. This paper takes advantage of multi-fractal model to detect fault in network traffic. Faults in a self-similar traffic destroy the singularity structure at the time points they occur, resulting in a significant deviation from those of normal traffic. For fault detection, we measure the degree of deviation of singularity exponent at every time segment through a deviation indicator Q based on structure function S_j(q). Since faults usually bring out abnormal bursts against natural ones in traffic, the proposed algorithm is thereby able to detect them. As demonstrated on simulated and real network traffic data, this algorithm can detect abnormal traffic loads with natural traffic bursts in the background.