An experimental design approach for link loss inference on large networks

An approach for link loss inference on large networks is proposed in this paper regarding with the cost of measurements. The measurements on large-scale networks usually cost much and the diagnosis of bottleneck on these networks are expensive and inefficient. We adapt Bayesian experimental design for measurement-path selecting with the total cost controlled in the budget to solve this problem. Through Bayesian experimental design with cost constrain, we can maximum the information getting from network measurements within the limited cost. Then the network inference method is used to infer the bottleneck link. The inference problem can be converted into a Linear Programming (LP) problem, which can be solved efficiently and accurately. We also carry out experiments to compare our approach with other approaches. The results show that with the same cost constrain, our approach is more accurate and has better performance.