Non-Stationarity and Multi-Scaling Behaviors in TCP Flow Arrivals

Scaling is a general term meaning that no particular scale controls the process under study. The literature on network traffic widely agrees with a mono-scaling Gaussian self-similar traffic model at time scales larger than the round-trip time (RTT) of TCP segments in the past ten years. In this paper, an extensive wavelet analysis is performed to the TCP flows at the small-time scales. The results show that the TCP flows exhibit a non-stationary Poisson behavior at timescales typically larger than a few hundred milliseconds. In order to identify the nonstationary- causing factors at the small-time scale, the TCP flow is analyzed along two dimensions--flow byte contribution and flow density. The results show that from the 'local' view, the dense flow is regarded as the main non-stationary causing factor but from the 'global' view, the alpha flow is regarded as the nonstationary causing factor.