Performance comparison of forecasting models applied to LAN/MAN traffic prediction

Abstract

The literature of network traffic analysis has successfully investigated several sophisticated models to be used in computer network traffic forecasting. Although these models have shown very good results in many controlled studies, the complexity of their implementation may be an important factor for preventing their large adoption in real production environment. We advocate that simpler forecasting models can also show very good accuracy — similar to the complex ones — for real scenarios of LAN/MAN network traffic, and being less intricate to implement and deploy in practical network management applications. In this paper we investigate the goodness of fit of nine classic forecast models applied to IP traffic samples drawn from real networks. The obtained results support our hypothesis given that the simpler investigated models demonstrate prediction accuracy very close to the advanced models for the studied scenarios.