The vision of self-driving networks integrates network measurements with network control. Processing data for each of the tasks comprising network control separately might be prohibitive due to the large volume and waste of computational resources. In this work we make the case of using the Weighted Stochastic Block Model (WSBM), a probabilistic model, to learn a task independent representation. In particular, we consider a case study of real-world IP-to-IP communication. The learned representation provides higher level-features for traffic engineering, anomaly detection, or other tasks, and reduces their computational effort. We find that the WSBM is able to accurately model traffic and structure of communication in the considered trace.

Top