IP/MPLS network modeling using Bayesian networks to improve double failure recovery

Protocols engineering of IP/MPLS networks are constantly improving with new separated features and new resilience mechanisms. In the transportation of audiovisual signals domain we must compose with multicast protocols which are designed from other scientific developments. This audiovisual traffic due to his non springy nature presents a very huge sensitivity to network recovery after a failure and these effects can be amplified by end devices (encoding, decoding and MPEG IP encapsulation). In this way when we choose between engineering solutions the unique criterion of availability is not enough, we must complete by an impact analysis on the service made by the network resilience technics. In this paper, we propose a first approach to analyze the behavior of different protocols engineering to improve selection. We propose using Bayesian networks to compare performance on different criteria and we will illustrate with two engineering models. The results focus on a real improvement of availability by choosing the adapted engineering solution.