Application classification in cellular backhaul networks

Cellular backhaul networks carry user's traffic within encrypted tunnels. With tunneled traffic, the application classifier cannot identify the underline TCP/UDP connections and cannot use any IP packet header information. Yet, some statistical properties can be considered in this case too, such as: volume per tunnel, tunnel durations, inter-packets delays (however, sequential packets might belong to different TCP connections), packet sizes, and packet direction. This research deals with blind classification of the dominated application of user tunneled traffic into 4 classes: (1) Control, VoIP and IM; (2) Video streaming; (3) Web browsing; and (4) file download/sharing. Our experimental results show that such a blind on-line classifier can classify the dominated application of a user tunnel into these four classes with more than 80% accuracy.