Improved Locality-Aware Grouping in Overlay Networks

The performance of peer-to-peer and overlay networks depends to a great extent on their awareness of the underlying network’s properties. Several schemes for estimating end-to-end network distances have been proposed to simplify this task. The mOverlay framework identifies groups of nodes that are near to each other in the network topology. Instead of distances between nodes mOverlay measures distances between groups. However, mOverlay’s locating procedure has a number of drawbacks. We propose an alternate method for identifying groups using Meridian’s closest node search. Simulation results based on PlanetLab measurements indicate that the Meridian-based approach is able to outperform mOverlay in terms of joining delay, the size of the identified groups, and their suitability for a distance estimation service.