On Real-time Failure Localization via Instance Correlation in Optical Transport Networks

Failure localization serves as a key to an effective fault management plane in the Internet backbone. This paper investigates a novel failure localization approach, namely Instance Correlation based Fault Diagnosis (IC-FD), for achieving efficient fault management in Optical Transport Networks (OTN). The IC-FD is aimed at real-time localization of failed components in the optical layer of OTN through correlation of alarms and status changes of network devices (referred to as instances) via a learned binary classifier. The outcome of IC-FD is one or multiple instance correlation trees (ICT) where the instances corresponding to the faulty network devices are taken as the tree roots. Notably, the proposed binary classifier is characterized by an intelligent feature extraction of historical instance correlation in dimensions of time, board/alarm attribute, network topology, and traffic distribution. Extensive case studies are conducted to demonstrate the advantages gained by IC-FD in terms of its high precision and low computation complexity, as well as analysis of its performance due to various environmental turbulence such as network topology, traffic diversity and noise alarms.