ANALYSIS OF NON-STATIONARY QUASI-PERIODIC INTERNET TRAFFIC BY THE METHOD OF INSTANT SPECTRA

Abstract

At large aggregation points, traffic growth processes are observed, which can be modeled by non-stationary quasi-periodic functions. A formal and informal analysis of traffic at a number of international aggregation points was performed. The traffic growth model was confirmed in the form of an algebraic sum of the main regular component and a number of cyclic components. The main cycles detected are in good agreement with the laws of rotation and rotation of the Earth and its interaction with neighboring celestial bodies. Not entirely obvious, but confirmed by the actual measurements of traffic, the dependence of human biological and social cycles on the rotation of all the same stars turned out to be. The daily, weekly, seasonal and annual traffic cycles turned out to be clearly pronounced. The nine-month and four-month cycles were unexpected. The latter are reflected only in the ancient calendars of the Slavs. In the daily dynamics, natural cycles prevail: day-night, the division of days into business and home traffic. Definitely unexpected was the discovered effect of desynchronization of Ukrainian traffic relative to the traffic of other European countries. The article concludes that this is due to incorrect determination of the decree time. A number of traffic dynamics models are proposed for solving subsequent prediction problems. It was found that in short time intervals the amplitude and phase components of the cyclic components are almost constant. An algorithm for selecting amplitudes of cyclical components and a regular trend component in a polynomial model of the latter is given. Software has been developed for solving similar problems of analyzing and predicting traffic using the instant spectra method.