A Method for Scalable Real-Time Network Performance Baselining, Anomaly Detection, and Forecasting

Abstract

Communication is the lifeblood of any business. Today, communication is predominantly facilitated by digital packets transported over the interconnected arteries of the data network infrastructure. It is imperative that this infrastructure is well managed, that unexpected behavior is quickly identified and explained, and that problems are predicted and preempted. Therefore, network performance management systems should be able to detect unusual or anomalous behavior as it happens, and quickly trigger automatic analysis or alert a human operator. Growth trends in network traffic must also be identified so that future problems may be anticipated and prevented. To meet these challenges, this paper proposes an integrated, scalable method to perform baselining, anomaly detection, and forecasting on time series network metrics. The method is based on the popular Holt-Winters triple exponential smoothing technique – a technique that compares favorably to other more complex and costly approaches.