Analytic models and characteristics of video traffic in high speed networks

Empirical studies suggest that video traffic streams in high speed BISDN/ATM telecommunication networks can be modeled as ARMA processes, but some results appear to be inconsistent due to differences in scene content and in the granularity of the time series. We propose a generic model for single source ARMA/ARIMA video streams that explains the effects of aggregation and sampling frequency at different timescales. We extend this to multiplexed independent video streams and describe how specific empirical results fit within the framework. We show that realizations of some models are non-stationary, and that restrictive assumptions of stationarity can be relaxed in the general case. Empirical results are consistent if the timescale is considered. The superposition of multiple independent video streams has certain predictable and consistent ARMA/ARIMA characteristics.<>