Statistical Error Analysis on Recording LRD Traffic Time Series

摘要

Measurement of LRD traffic time series is the first stage to experimental research of traffic patterns. From a view of measurement, if the length of a measured series is too short, an estimate of a specific objective (e.g., autocorrelation function) may not achieve a given accuracy. On the other hand, if a measured series is over-long, it will be too much for storage space and cost too much computation time. Thus, a meaningful issue in measurement is how to determine the record length of an LRD traffic series with a given degree of accuracy of the estimate of interest. In this paper, we present a formula for requiring the record length of LRD traffic series according to a given bound of accuracy of autocorrelation function estimation of fractional Gaussian noise and a given value of H. Further, we apply our approach to assessing some widely used traces in the traffic research, giving a theoretical evaluation of those traces from a view of statistical error analysis.