Semi-empirical description and projections of Internet traffic trends using a hyperbolic compound annual growth rate

For the purpose of projecting future macroscopic Internet traffic, we have analyzed historical U.S. and global Internet traffic volumes previously reported for 1990 to the present. The empirical long-term traffic trends are characterized by growth rates that have generally been decreasing over time, and so are not adequately described as exponential growth. Based on regression analyses, we also conclude that the decreasing growth rates are not well represented by the saturation characteristic of the logistic function alone. However, we find the observed traffic data are well reproduced over ranges spanning more than six orders of magnitude of traffic volume and a period of 17 years by a hyperbolic dependence of the compound annual growth rate on time. Using a semi-empirical hyperbolic function to model the growth rate we have carried out linearized regression analyses of a combination of historical and near-term traffic forecasts to project macroscopic Internet traffic to 2020 for several geo-economic regions, market segments, and application categories. In contrast to the decade of 2000-2010 when global Internet traffic grew more than 100-fold, the projections indicate that over the decade 2010-2020 the global wireline Internet traffic will grow approximately 16 times larger to approach 250 exabytes per month and the global mobile Internet traffic will grow approximately 150 times larger to approach 40 exabytes per month. The projections also indicate that over the decade the compound annual growth rate of global wireline Internet traffic will decrease from approximately 45 percent to 25 percent per year and the growth rate of global mobility Internet traffic will decrease from approximately 170 percent to 30 percent per year. By considering both the historical trends and near-term forecasts, our analyses and projections call attention to those traffic categories for which estimates of future Internet traffic may carry lesser or greater uncertainty.