Revisiting multifractality of TCP traffic using multifractal detrended fluctuation analysis

Understanding the characteristics of network traffic is critical to many network design and engineering problems, such as router buffer dimensioning and delay-sensitive service provisioning in the Internet backbone. We re-analyze the multifractal features of TCP traffic collections spanning 19 years by means of multifractal detrended fluctuation analysis. Through multifractal analysis, the generalized Hurst exponents and scaling exponents are derived. These results indicate that the TCP traffic is multifractal. The widths of the singularity spectrum show that the multifractal features of TCP traffic are weakened with increasing bandwidth and statistical multiplexing, but cannot be eliminated. In particular, comparing the generalized Hurst exponent of the original TCP traffic series with the shuffled and surrogate series, we conclude that the multifractal nature of TCP traffic is due to both long-range correlation and fat-tailed probability distributions. Moreover, the results show that the multifractality due to the correlation has a greater contribution than that from a fat-tailed probability distribution.