Investigation of the performance of GAFT, a novel network anomaly fault detection system

We investigate the performance of a novel hierarchical, distributed, multi-tier, multi-hierarchical, distributed, multi-tier, multi-window fault detection system, namely the Generalized Anomaly and Fault Threshold (GAFT) system. We have carried out extensive simulation experiments of network resource and service deterioration, under two kinds of challenging conditions: in the presence of increasing intensities of class-alien traffic congestion and for a fixed intensity of anomaly traffic, but of varying characteristics, i.e., the distribution of fault packet length and inter-arrival interval. The corresponding numerical results demonstrate that GAFT is very efficient and can reliably detect the soft fault with traffic anomaly intensity as low as three percent of the typical background traffic intensity. Moreover, while GAFT is sensitive to the characteristics of the distributions that shape the fault under observation, it is very powerful in discerning even the most challenging cases, when the fault has very similar defining distribution characteristics to the prevailing background.