Internet Protocol Packet Delay Variation measurements in communication networks: How to evaluate measurement uncertainty?

Measurement of Quality of Service (QoS) parameters is a relevant task in communication networks. Similarly to other contexts, the measurement result has to be completed with the related uncertainty in order to provide information about the significance and reliability of the overall measurement process. This aspect is particularly crucial in communication networks. In fact, the measured values of QoS parameters have often to be compared against suitable thresholds to check their compliance with required quality constraints (QoS levels delivered to the user are generally offered for a fee) or to reliably verify the effect of new services conveyed by the network on its performance. Nevertheless, uncertainty estimation in QoS parameters measurement is still an open issue. Most literature concerning communication and computer networks test and measurement clearly evidences the lack of the advisable practice of uncertainty evaluation in QoS parameters measurement according to the guide for the expression of uncertainty in measurements (GUM). In addition, it is worth emphasizing that uncertainty modeling and evaluation in QoS parameters measurement are not trivial tasks because several aspects are involved in a QoS measurement process. All these aspects can be thought of as sources of measurement uncertainty, capable of significantly influencing the reliability of measurement results. The paper investigates these issues by promoting and assessing the applicability of GUM to the specific and critical context of communication and computer networks test and measurement. More specifically, a simple model for uncertainty evaluation in Internet Protocol Packet Delay Variation measurement is proposed: a fundamental QoS parameter in Voice over IP and real time video streaming applications. The model is based on two main contributions: the former accounting for the instrumental measurement uncertainty and the latter for the measurand variability. A number of tests in a simulated scenario proves the effectiveness of the proposed model. To exemplify its practical application, some experiments in a real test-bed, i.e. the local area network of the University of Cassino and Southern Lazio, are finally conducted.