No abstract

No abstract

No abstract

No abstract

SUMMARYThe design of anomaly detection (AD) methods for network traffic has been intensively investigated by the research community in recent years. However, less attention has been devoted to the issues which eventually arise when deploying such tools in a real operational context. We designed a statistical based change detection algorithm for identifying deviations in distribution time series. The proposed method has been applied to the analysis of a large dataset from an operational 3G mobile network, in the perspective of the adoption of such a tool in production. Our algorithm is designed to cope with the marked non-stationarity and daily/weekly seasonality that characterize the traffic mix in a large public network. Several practical issues emerged during the study, including the need to handle incompleteness of the collected data, the difficulty in drilling down the cause of certain alarms, and the need for human assistance in resetting the algorithm after a persistent change in network configuration (e.g. a capacity upgrade). We report on our practical experience, highlighting the key lessons learned and the hands-on experience gained from such an analysis. Finally, we propose a novel methodology based on semisynthetic traces for tuning and performance assessment of the proposed AD algorithm.