Forecasting and Event Detection in Internet Resource Dynamics using Time Series Models

Abstract: At present Internet has emerged as a country's predominant and viable data communication infrastructure. The Autonomous System (AS) resources which are building blocks of the Internet are AS numbers, IPv4 and IPv6 Prefixes. AS number growth is one of Internet infrastructure development indicators. Hence understanding on long term trend and stochastic variation behaviour are essential to detect significant events during the growth. In this work, time series based approximation is considered for mathematical modelling and forecast the yearly AS growth. The AS data of five countries namely India, China, Japan, South Korea and Taiwan are extracted from APNIC archive. ARIMA models with different Auto Regressive and Moving Average parameters are identified for forecasting. Model validation, parameter estimation, point forecast and prediction intervals with 95 % confidence levels for the five countries are reported in the paper. The significant level change in variations, positive growth percentage in Inter Annual Absolute Variations (IAAV) and higher percentage of advertised ASes when compared to other countries indicate India's fast growth and wider global reachability of Internet infrastructure from 2007 onwards. The correlation between IAAV change point and GDP growth period indicates that service sector industry growth is the driving force behind significant yearly changes.