Availability estimation of optical network links using a Bayesian model

Abstract

Common daily activities rely on information exchange and processing. Optical fiber links are the predominant way for transmitting vast amounts of data, thus enabling processing in remote data centers. As critical use cases like autonomous driving and biomedical procedures start relying on such infrastructure, the system’s availability becomes even more relevant. Assessing the availability of an optical link is a well-known problem, but inconclusive nonetheless. Finding the true link availability requires a perfect understanding of the complete underlying system, which is impossible to capture to such an extent. Hence, different approaches or models arise as we focus on approximating the true value. Here, we develop a hierarchical Bayesian model and compare it to various baselines. We show that the estimation methods present different behavior for separate scenarios. Moreover, a use case is investigated where services with varying availability requirements must be deployed. Using a Bayesian model to estimate the link availabilities produces, on average, the best accuracy among the considered baselines and provides worthy uncertainty estimations. Such estimations increase the network operator’s trust and allow more decision-making flexibility.