Network Link-Level Loss Inference Based On End-to-End Measurement

Internet is a massive, distributed network which continues to explore in size as e-commerce grows. The characteristics of a network include loss ratio, delays, etc.. It is essential to have link-level performance data for understanding the features of a network. Monitoring can be performed internally, but it is impractical to directly measure traffic characteristics at all internal devices for some reasons. A reasonable way is to infer the internal network behavior by "external" end-to-end measurements. This paper introduces network tomography, a field which we believe will benefit greatly from signal processing theory and algorithms.This paper focuses on the link-level loss ratio estimation based on end-to-end measurements.The paper proposes a method using the dependence tree model and upward-downward algorithm to infer link loss ratio. This method is suitable to multicast network. Multicast probe traffic is added from the root node on the ongoing traffic. Analyses are done at several receivers based on the received probes and the link loss ratios are inferred. The key to this approach is that multicast traffic introduces correlation in the end-to-end losses measured by receivers. This correlation can, in turn, be used to infer the loss behavior of the links within the multicast routing tree spanning the sender and receivers. This enables the identification of links with higher loss rates as candidates for the origin of the degradation of end-to-end performance. Also it introduces unicast related algorithm.The paper also introduces the software's topology discovery module. The module is based on ping and traceroute network tool. The display layer will invoke the topology information stored in the database of topology.