Multiscale Internet Statistics: Unveiling the Hidden Behavior

摘要: Being able to characterize and predict the behavior of Internet users based only on layer 2 statistics can be very important for network managers and/or operators. Operators perform a low level monitoring communications at entry points, independently data encryption even without being associated with itself. Based this data, it is possible optimize access service, offer new security threats detection services infer behavior, which consists identifying underlying web application that responsible by traffic different time instants usage dynamics applications. Several identification methodologies have been proposed over years classify identify IP applications, each one having its own advantages drawbacks: port-based analysis, deep packet inspection, behavior-based approaches, learning theory, among others. Although some them are efficient when applied specific scenarios, all approaches fail available or under restrictions. In work, we propose use multiscaling characteristics differentiate applications Markovian model user actions time. By applying methodology Wi-Fi generated accessing common services/contents through HTTP (namely social networking, news web-mail applications), was achieve good prediction behaviors. The classification results obtained show developed has potential efficiently identify, behaviors statistics.