Detection and quantification of self- similarity in data traffic for prediction of performance of marine data file transfer

The time-sensitivity of large marine data files over a communication network necessitates accurate simulative prediction of the file transfer performance. Careful data traffic modelling is required to fit the actual traffic characteristics for subsequent generation of synthetic traffic traces and feeding them into a simulation model. Classical models have recently proved inadequate due to the discovery of self-similarity (fractal behaviour) in data traffic. This paper attempts to systematise the mathematical background of self-similarity and the ways it manifests itself in stochastic processes modelling data traffic. Relevance of self-similarity to traffic description and measurements is discussed. Results of a research effort at the Department of Marine Electronics of the Maritime Institute in Gdansk are described, which focus on the development of a software tool for detection and quantification of self- similarity in observed or synthetically generated data traffic.