Network coding for loss tomography

This paper addresses the problem of monitoring networks for congestion by making active end-to-end measurements with probe sequences over overlay networks. Traditionally, link loss rates are inferred by sending probe packets from a set of sources at the network edge to a set of receivers over a mesh of multicast trees. We first approached this problem in [3] where we showed that enabling internal nodes of the overlay with network coding capability delivers increased bandwidth efficiency and estimation accuracy when compared to multicast-based tomography. Taking our initial work in [3]¹ as a starting point, in this paper we develop sophisticated link loss estimators and present a wide range of experimental results. These estimators are distributed, low-complexity link-loss inference heuristics that can be applied to arbitrary network topologies.