Fluctuation of the download network

TL;DR

This paper investigates the fluctuation behavior of a download network, revealing a variable power-law scaling between fluctuation and average download rate, influenced by external factors and congestion, challenging previous universal fluctuation class assumptions.

Contribution

It demonstrates that the fluctuation exponent varies with the time window, providing a new understanding of network traffic fluctuations beyond prior universal classes.

Findings

Fluctuation exponent $eta$ varies from 1/2 to 1 with time window size.

Power-law scaling exists between fluctuation and average flux.

Crossover behavior explained by external driving fluctuation model.

Abstract

The scaling behavior of fluctuation for a download network which we have investigated a few years ago based upon Zhang's Encophysics web page has been presented. A power law scaling, namely $\sigma \sim < f> ^ \alpha $ exists between the dispersion $\sigma$ and average flux $<f>$ of the download rates. The fluctuation exponent $\alpha$ is neither 1/2 nor 1 which was claimed as two universal fluctuation classes in previous publication, instead it varies from 1/2 to 1 with the time window in which the download data were accumulated. The crossover behavior of fluctuation exponents can be qualitatively understood by the external driving fluctuation model for a small-size system or a network traffic model which suggests congestion as the origin.