This work presents a distance-based agglomerative clustering algorithm for inferential monitoring based on end-to-end measurements obtained at the network edge. In particular, we extend the bottom-up clustering method by incorporating the use of nearest neighbors (NN) chains and reciprocal nearest neighbors (RNNs) to infer the topology of the examined network (i.e., the logical and the physical routing trees) and estimate the link performance characteristics (i.e., loss rate and jitter). Going beyond network monitoring in itself, we design and implement a tangible application of the proposed algorithm that combines network tomography with change point analysis to realize performance anomaly detection. The experimental validation of our ideas takes place in a fully controlled large-scale testbed over bare-metal hardware. View this paper.