Applying a recurrence plot scheme to analyze non-stationary transition patterns of IP-network traffic

This paper describes a recurrence plot (RP) approach to the analysis of non-stationary transition patterns of IP-network traffic. To get a quantitative measure of dynamical transition patterns of IP-network traffic, we used the values of determinism (DET) defined by the recurrence quantification analysis (RQA). Also, when evaluating the fractal-based properties of IP-network traffic, we focused on two parameters: (i) the long-range dependence (LRD)-related scaling parameter α derived from the detrended fluctuation analysis (DFA) and (ii) the range of the generalized fractal dimension. In applying this method to traffic analysis, we performed two kinds of traffic measurement in Tokyo, Japan, and derived the values of DET and fractal-based parameters of IP-network traffic over time. In checking the measured network traffic, we found that the characteristic with respect to DET and self-similarity seen in the measured network traffic fluctuated over time, with different time-variation patterns for two measurement locations. Results also confirmed that a larger value of DET or accumulated DET reflected increases in the degree of LRD of IP-network traffic and that the accumulated DET reflected the decreases in the degree of multi-fractality of IP-network traffic. As a result, we confirmed that RP-based measures can be effective for evaluating the non-stationary transition patterns of IP-network traffic in terms of quantitative fractal-based properties.