3D-LD: a Graphical Wavelet-based Tool for Analyzing Non-stationary and Scaling Processes
Steve Uhlig · 2003
This paper proposes a novel graphical tool based on wavelets for studying the nonstationary and scaling properties of general signals, which we call the 3D-LD. We show that it allows to better study the timescale-dependent qualitative properties of non-stationary and scaling signals, compared to the classical techniques. Based on the 3D-LD, this paper analyzes several long network traffic traces spanning the years from 1993 to 2001, to understand the actual nature of scaling in TCP flows arrivals. The study of these traces shows that the TCP flows arrivals is a complex scaling process, which is highly non-stationarity from several point of views. Based on this analysis, we propose an analytical model for the dynamics of the TCP flows arrivals process, which takes into account all timescales ranging from seconds to hours. Finally, we show that simulating realistic TCP flows arrivals conforming to this model is feasible.