Behavior of R/S Statistic Implementations under Time-Domain Operations

Complex correlation structure of current communications network traffic requires the use of algorithms for detecting the degree of dependence in a sample path of the network traffic process. Several algorithms for accomplishing the above exists with varying degrees of accuracy and convergence. In this paper we study the R/S statistic, in particular, the accuracy, convergence and its effectiveness under time aggregation for several long-memory characteristics (fGn and FARIMA). We present the selection of the optimal values of the low and high-ends in the linear fit for three implementations of the R/S statistic. Based on the above we show that different implementations give varying degrees of accuracy, convergence and effectiveness under time aggregation