A practical approach for network fault detection

Abstract

Today's fault detection in commercial networks is still done in an inefficient way with alarms and threshold violations that treat measured indexes separately. To provide a practical network fault detection approach for the actual network operation, we investigate many actual network fault occurrences, and discover that many network faults can be characterized by unbalanced variation among measured network indexes that occur prior to or during an network fault occurrence. According to this general model of network fault, we propose a practical method for network fault detection that automatically extracts the unbalanced variation among measured indexes without the necessity of recognizing the physical meaning of them. Our evaluation shows that the proposed approach is applicable for the majority of measured indexes in the commercial networks, is efficient and scalable in performance, and with acceptable detection accuracy.