EARLY FAULT DETECTION IN SOFTWARE DEFINED NETWORKS

Abstract

Detecting faults early in network and communication systems is one of important functionalities of fault management. As these systems become large in size and scalability, complex in architecture and design, and dynamic in operation and control, detecting faults become even more difficult and challenging. This study presents an method for solving this problem in emerging and promising software defined networks (SDN) with above characteristics. This method collects event log data from switches and uses machine learning techniques to detect faults. The method is associated with the SDN controller to monitor, analyze and notify faults to the system operator through applications. Evaluating the method includes the extensibility of the controller using the Ryu open source tool and several experiments on the Spark event log dataset using the Random Forest technique.