Publication Date


Degree Type

Master's Project

Degree Name

Master of Science (MS)


Computer Science


End-to-End Available Bandwidth (AB) is a real-time network metric that is useful for a wide range of applications including content distribution networks, multimedia streaming applications and overlay networks. In a large network with several thousand nodes, it is infeasible to perform all-pair bandwidth measurements as AB measurements could induce traffic overhead along the path. Also because of its dynamic nature, the measurements have to be performed frequently thus imposing significant probe traffic overhead on the network. In this paper, we discuss a clustering based distributed algorithm to infer the AB between any pair of nodes in a large network based on measurements performed on a subset of end-to-end paths. The algorithm was validated on Planet-Lab and for some nodes, 80% of the inferences were within 50% of the actual value.