Stationarity in OWD time series

IT is a widely accepted view that Internet tra c and thus the related quantities show long-range dependence (LRD). Besides other features, in practice, this means slowly decaying correlations, often according to some power law: C(τ) ∼ τ−α. Recent results suggest that this phenomenon is valid only on certain time scales, depending on the actual measurement. Stationary periods were detected in various types of Internet related time series with di erent characteristic time scales, e.g. loss rates (1 s 1 min) or TCP transfers (1 hour) [1]. Note that these observations depend on the sampling rate and the total length of measurement as well. In this contribution we study statistical properties of one-way delay (OWD) time series, focusing on the question of stationarity. Understanding the nature of stationarity is relevant for both modelling and the development of more optimized network algorithms, since one can infer the network state from such data.