A Real-Time Identification System for VoIP Traffic in Large-Scale Networks

Recently, VoIP applications win most of the users' favor with their high service quality and low price. However, owing to the flexible of VoIP services, they will cause incalculable damage if these services are used improperly. In order to make VoIP applications serve humans better, it is important to keep VoIP applications under supervision. The upgrading of VoIP technology makes the traditional identification methods inefficient, it becomes more difficult to identify VoIP traffic. For the challenge of identifying VoIP traffic, the deep learning technology provides us a new solution. With supervised learning method, we can build feature set for identifying VoIP traffic. This way is not only able to find the most useful features, but also it can get rid of human effort in exploring features. In this paper, we adopt CLNN(Convolutional Neural Networks, Long Short-Term Memory) to extract features for accurate VoIP application identification. In addition, we design a real-time identification system to capture VoIP traffic in a large-scale network and identify their application types with the features we trained. The evaluation results verify that our system can identify VoIP traffic timely and accurately.