A closer look at ATM traffic measurements and their statistical analysis

This paper presents an analysis of cell-level ATM traffic measurements collected from Sprint´s network through a Telcordia Technologies (formerly Bellcore) custom-built ATM traffic recording device during a week in the summer of 1998. Analysis of the total traffic during one busy hour on one OC-3c link strongly indicated that the total aggregate traffic was self-similar with long-range dependent characteristics, and could be adequately modeled by the fractional Brownian motion (FBM) traffic model for time scales of current performance and engineering interest. Analysis on the finer time scale (100 microseconds) characteristics of the aggregate ATM traffic does not support a multifractal hypothesis. Simple (mono-fractal) scaling holds from 100 microseconds to a few hundred seconds. However the marginal distribution at time scales of 100 microseconds is no longer Gaussian, which can invalidate traffic performance predictions from the FBM model at that time scale. Beyond the aggregate traffic level, additional traffic characteristics were investigated. The three classes of service offered were CBR, VBR, and UBR; most of the subscriptions were for VBR service. The traffic was primarily type AAL 5. The class-of-service was not directly indicative of AAL type. The traffic patterns generated by customers had more to do with the AAL type and applications than with service subscription. The following observations drawn regarding the traffic characteristics by AAL type are presented