Architecture of Generalized Network Service Anomaly and Fault Thresholds

Zheng Zhang¹, Constantine Manikopoulos¹, and Jay Jorgenson²

¹ECE Department,
New Jersey Institute of Technology,
University Heights,
Newark, NJ 07102, USA
zxz9622@njit.edu,
manikopoulos@adm.njit.edu

²Department of Mathematics,
CCNY, Convent Ave. at 138 ST.,
New York, NY 100031, USA
jjorgenson@mindspring.com

Abstract. In this paper we introduce GAFT (Generalized Anomaly and Fault Threshold), featuring a novel system architecture that is capable of setting, monitoring and detecting generalized thresholds and soft faults proactively and adaptively. GAFT monitors many network parameters simultaneously, analyzes statistically their performance, combines intelligently the individual decisions and derives an integrated result of compliance for each service class. We have carried out simulation experiments of network resource and service deterioration, when increasingly congested in the presence of class-alien traffic, where GAFT combines intelligently, using a neural network classifier, 12 monitored network performance parameter decisions into a unified result. To this end, we tested five different types of neural network classifiers: Perceptron, BP, PBH, Fuzzy ARTMAP, and RBF. Our results indicate that BP and PBH provide more effective classification than the other neural networks. We also stress tested the entire system, which showed that GAFT can reliably detect class-alien traffic with intensity as low as five to ten percent of typical service class traffic.

LNCS 2216, p. 241 ff.

Full article in PDF (130 KB)

Online publication: October 10, 2001
helpdesk@link.springer.de
© Springer-Verlag Berlin Heidelberg 2001