Analysis of Network Distance Prediction with Global Network Positioning

Coordinates-based mechanisms have proven to be useful in a peer-to-peer architecture to predict Internet network distance, even more so than the already existing IDMaps scheme. One such mechanism has been analyzed known as Global Network Positioning (GNP) to provide insight on a current state of the art technology in network distance prediction. GNP instantiates a virtual geometric space by applying coordinates to nodes in a network; coordinates are computed by utilizing an objective function. A set of landmark hosts are first deployed into the geometric space to implement a set of reference points for any newly discovered host. Hosts also maintain their own coordinates, making it accessible to retrieve inter-host network distance upon discovering eachother by utilizing a distance function. The variables associated with GNP must be tweaked to maximize efficiency. Through performed experiments using PlanetLab [1], a service that allows access to hosts at multiple research institutions across the globe, the factors that affect GNPs performance are analyzed.