Higher-Order Correlations in Data Network Traffic

In this study, we investigate higher-order correlations in data network traffic. We, in particular, study the statistical properties of the volatility, a well-established concept in financial market analysis, of two sets of Ethernet LAN data network traffic. We find that the volatility follows a log-normal distribution regardless of the window size. The volatility is further analyzed by using a detrended fluctuation analysis (DFA), a method especially suitable for testing the long-range correlation of non-stationary data. With a DFA, we find that the volatility exhibits either a long-range dependence or 1/f noise, depending on the window size. Although the two sets of time series obtained from corresponding LAN traces show similar long-range correlations, their volatilities disclose somewhat different statistical properties.