Parameter estimation in FARIMA processes with applications to network traffic modeling

Traffic measurements in many network environments demonstrate the coexistence of both long- and short-range dependence in traffic traces. In this paper, we use the fractionally integrated autoregressive moving average (FARIMA) processes with non-Gaussian innovations to describe packet arrival rate in unit time. Specifically, we investigate cepstrum-based approaches for parameter estimation in FARIMA processes. We examine the fractional differencing parameter estimation procedure based on the smoothed periodogram and the log spectrum. The simulation results demonstrate that the proposed cepstrum approach gives better estimation accuracy than the conventional least-square spectrum fit. Usefulness of the results presented is demonstrated on real network traffic traces by considering spectral fitting metrics.