Choosing Internet paths with high bulk transfer capacity
Jacob Strauss · 2002
Many applications wish to predict what bandwidth TCP can achieve on a path between Internet hosts without performing extensive measurements. More precisely, they would like to know the bulk transfer capacity (BTC) of a given path. This thesis compares the ability of several BTC estimation methods to predict which of a pair of Internet paths will have the higher measured BTC. Methods tested include TCP loss rate models, an available bandwidth measuring tool named Pathload [17], path capacity measurements, Asymptotic Dispersion Rate measurements, and a new approach we call Squeezed Pairs. We tested each estimation method on paths between hosts on the RON testbed [5]. When taking measurement costs into account, the best method we have found to pick the faster of a pair of paths determines the Asymptotic Dispersion Rate of each path. The ADR test consists of sending two streams of 15 packets back-to-back and calculating the average spacing between probe packets at the receiver. The ADR test correctly predicts which path will have a higher BTC in 80% of all path pairings.