Characterization and generation of synthetic data traces for IP traffic modeling

Abstract

fgN and fARIMA (0,d,0) are two of the models frequently employed in Internet traffic. However, there is not a set of data traces available. In this work, a set of 187 fGN data traces each one of 64KPoints were generated with three tools: S- Plus, SelQoS; and a fGN-MatlabScript; and a cumulative analysis with 1144 files was done. From this study, it is concluded that the best algorithms respect to the Minimum Mean Squared Error (MMSE) for Hurst index estimates were Modified Allan Variance in time domain far followed for R/S statistic, while in frequency domain Periodogram presented the MMSE followed very close for Local Whittle. A set of 53 "well-behaved" data traces with different theoretical values HT were obtained and will be publicly available. The fGN data traces generated by our tool SelQoS compared with those generated with S-Plus, fGN, fARIMA (0, d, 0), and fGN-MatlabScript showed the best performance relative to MSE. For all data generation methods, the higher H is, the bigger MSE is.