Network performance measurements have been broadly used in order to debug and to assess the reliability of the network. In general performance measurement is a resource consuming process, which in practice derives in hardly scalable systems. However, in order to control and monitor network resource usage and to assess the quality of multimedia traffic, it is broadly accepted that efficient and scalable network performance evaluation infrastructures must be present in nowadays networks. Nevertheless, most of current research is focused on the assessment of the different performance metrics in order to determine if the quality of the delivered service is correct. Differently, the main contribution presented in this paper, relies on the fact that the accurate estimation of the network metrics is generally not necessary, given that it is much more efficient to directly detect service disruptions. To this end, we propose an algorithm to detect anomalies on the service level delivered to the users. Our solution is based on the distance measurement between acquired reference distributions on the Inter-Packet Arrival Times. This simple to measure time-series permit to detect very efficiently QoS disruptions, making of the solution a very good candidate to be used on Service Level Agreement assessment systems, as we prove in our experimental set up, where we evaluate the performance and accuracy of the proposal in a real scenario.

Peer Reviewed