A Binary Independent Component Analysis Approach to Tree Topology Inference

Using multicast probes to infer network topologies and internal link/node characteristics is an attractive approach due to its bandwidth efficiency and suitability for large-scale measurements. In this paper, we propose a new approach to tree topologies inference by exploiting dependence among end-point receivers. We first show that under the assumption of independent failure of intermediate nodes or links, inferring tree topology is a special instance of the more general problem of binary independent component analysis (bICA), and thus is amiable to existing analytical results and algorithms for bICA. Then, we propose the seqBICA algorithm that is tailored for tree topology inference. Evaluation study shows that the proposed algorithm outperforms existing approaches in convergence speed and accuracy even when the number of measurements is small.