Hybrid Power-Law Models of Network Traffic

TL;DR

This paper introduces an enhanced probabilistic network model that combines preferential attachment with additional parameters to better capture observed phenomena in large-scale streaming network traffic data, improving theoretical understanding.

Contribution

It extends existing power-law network models by incorporating new components to account for leaves and unattached nodes, aligning theory more closely with empirical data.

Findings

Model accurately replicates observed network distributions

Incorporates parameters for leaves and unattached nodes

Provides a framework for analyzing large-scale network data

Abstract

The availability of large scale streaming network data has reinforced the ubiquity of power-law distributions in observations and enabled precision measurements of the distribution parameters. The increased accuracy of these measurements allows new underlying generative network models to be explored. The preferential attachment model is a natural starting point for these models. This work adds additional model components to account for observed phenomena in the distributions. In this model, preferential attachment is supplemented to provide a more accurate theoretical model of network traffic. Specifically, a probabilistic complex network model is proposed using preferential attachment as well as additional parameters to describe the newly observed prevalence of leaves and unattached nodes. Example distributions from this model are generated by considering random sampling of the networks created by the model in such a way that replicates the current data collection methods.