Solid end-to-end probing and analysis method for estimating network loss

Abstract

Overlay network monitoring is one of the most important issues in the design and operation of overlay networks. It faced with increasing demand for better throughput and response time performance. Given an overlay network with n end hosts, existing systems either require O(n2) measurements, and thus lack scalability, or can only estimate the latency but not failures or congestion. We propose a scalable deterministic network monitoring scheme that selectively monitors k linearly independent paths which can fully describe all the O(n2) paths. We use the loss rates and latency of the collected k path to deduce the loss rate and latency of the rest paths. Our method only assumes knowledge of the underlying IP topology, with links dynamically varying between normal and lossy. In this paper, we propose a novel path loss rate inference algorithm which improves, implements and extensively estimates such a monitoring method. We carry out our experiments in two different scenarios and the results from the experiments demonstrate that our approach increases accuracy in monitoring congested paths compared to other representative schemas'. Especially in a router-level topology, our approach can deliver almost a 12.5% increase.