Testing the Scalability of Overlay Routing Infrastructures

Recent studies have demonstrated the utility of alternate paths in improving connectivity of two end hosts. However studies that comprehensively evaluate the tradeoff between its effectiveness and overhead are lacking. In this paper we carefully characterize and evaluate the trade-off between (1) the efficacies of alternate path routing in improving end-to-end delay and loss, and (2) the overheads introduced by alternate routing methodology. This would help us to test the scalability of an overlay network. We collected ping data on PlanetLab and studied the above trade-off under different parameter settings such as path sampling frequency, overlayconnectivity, number of overlay hops etc. Results from ten sets of measurements using 35 to 36 of the PlanetLab nodes are used to test the effect of the various parameters. We find that changing epoch duration and connectivity helps reduce the overheads by a factor of 4 and 2 respectively at the cost of some lost performance gain.