Availability Estimation of Optical Network Links Using Multilevel Bayesian Modeling

Estimating availability in transport networks is essential to successfully provision connectivity services under various requirements. As modern Service Level Agreements (SLAs) are becoming more demanding, it is crucial to estimate the true availability of an end-to-end path as soon and reliably as possible. Network operators might have expert knowledge regarding the theoretical availability of the network components, or they might have some measured data to infer such information. However, in either case, the true steady-state availability is unknown and must be inferred as a combination of the above. Traditionally, the estimation of Quality of Service (QoS) metrics, like availability, is done by producing a scalar value as an approximation. This can often be misleading and an oversimplification. In this paper, we build a multilevel Bayesian model that provides a probabilistic estimate of all network links' availability by exploiting expert knowledge and the underlying data. The value of this methodology is greater for scenarios with scarce data, where the inference quality demonstrates remarkable stability regardless of the randomness of the incoming data. We demonstrate the effectiveness of our approach through simulation experiments and compare our results with the empirical interval availability. Our work has important implications for the design and management of optical networks since it constitutes a valuable tool that provides accurate availability estimations, which can be used for sophisticated decision-making.