Network Tomography via Sparse Bayesian Learning

Network tomography aims to estimate internal link states from end-to-end path measurements. In this letter, we propose a new network tomography scheme using sparse Bayesian learning (SBL). SBL takes advantage of the sparsity of link-level parameters and implements Bayes' rule based on the Gaussian Prior. Through the proposed scheme, most links' metric values are regarded as zero while the significant ones are identified. Therefore, it is especially useful when our goal is to locate the congested links. We conduct the simulation experiments of delay estimation and congestion detection, demonstrating the distinctive characteristic of detection rate and false discovery rate through the proposed scheme.