Seeking the best Internet Model

TL;DR

This paper evaluates nine Internet models using a multivariate statistical approach to compare multiple topological properties against real Internet data, revealing none fully replicate the Internet's complex structure.

Contribution

It introduces a multivariate statistical methodology to compare Internet models across numerous topological metrics, highlighting the limitations of existing models.

Findings

None of the models accurately reproduce Internet topology.

The methodology considers 21 network measurements.

The approach provides a comprehensive comparison framework.

Abstract

The models of the Internet reported in the literature are mainly aimed at reproducing the scale-free structure, the high clustering coefficient and the small world effects found in the real Internet, while other important properties (e.g. related to centrality and hierarchical measurements) are not considered. For a better characterization and modeling of such network, a larger number of topological properties must be considered. In this work, we present a sound multivariate statistical approach, including feature spaces and multivariate statistical analysis (especially canonical projections), in order to characterize several Internet models while considering a larger set of relevant measurements. We apply such a methodology to determine, among nine complex networks models, which are those most compatible with the real Internet data (on the autonomous systems level) considering a set of 21 network measurements. We conclude that none of the considered models can reproduce the Internet topology with high accuracy.