Network Bandwidth Usage Forecast in Content Delivery Networks

Abstract

Operational burden of a Content Delivery Network that is a vast overlay network on top of current Internet Architecture can be alleviated by forecasting Content Delivery Network bandwidths. The purpose of this paper is to forecast network bandwidth usage for Content Delivery Networks’ Points of Presence. In this paper we compare Seasonal Auto-Regressive Integrated Moving Averages and Artificial Neural Networks that can be used for predicting and minimizing operational costs of Content Delivery Networks via resource allocation, server allotment and local ISP bandwidth contract costs. We directly forecast end-user to Content Delivery Network bandwidth, so it can directly be used to lower end-user latencies. In this paper; we first conduct Self-Similarity Analysis and then utilize Seasonal Auto-Regressive Integrated Moving Averages and Artificial Neural Networks to predict bandwidth usage with 6.338% error.