Data-driven Evaluation of Anticipatory Networking in LTE Networks

Abstract

Anticipatory networking is a recent branch of network optimization based on prediction of the system state. Our work specifically tackles prediction-driven resource allocation for mobile networks. While some anticipatory networking concepts have been proposed in the literature, understanding of the potential real world gains is so far very limited. Future mobile networks will likely integrate such mechanisms, and thus it is of paramount importance to understand the actual performance improvements and in which scenarios they can be realized. Analyzing a month of LTE control channel information collected in four locations, we show how anticipatory networking can enhance current LTE networks. First, we propose a comprehensive optimization framework encompassing different forecasting solutions. Then, we provide a thorough analysis of the aggregated network traffic and the contributions of individual users. In particular, we show that predictable traffic accounts for more than 95% of the total traffic volume and that simple prediction and optimization techniques allow network operators to save 50% of the resources and/or on average more than double the offered data rate in our data set.