Analysis of Network Distance Prediction with Global Network Positioning Mathieu Rodrigue Department of Computer Science University of Hartford 17/09/2018
Background “Predicting Internet Network Distance With Coordinates-Based Approaches” Triangulated Heuristic IDMaps Global Network Positioning (GNP)
IDMaps Topological map of the internet Query HOPS Servers to get distance Consistent, but, many over predictions
Global Network Positioning Reference points are disseminated known as “Landmarks” Landmarks and hosts compute coordinates relative to each other by utilizing an objective function
Minimization Objective function for Landmarks:
Minimization Objective function for ordinary hosts:
Global Network Positioning Coordinates are applied to nodes within the network A standard vector distance function is utilized to predict network distance
Factors that affect results: Methodology Factors that affect results: Number of dimensions Number of Landmarks Network speed when sending ICMP ping packets Directional relative error:
Using the Planet Lab service: Methodology Using the Planet Lab service: Choose five initial international reference points to serve as landmarks/tracers Binghamton, New York Hamburg, Germany Christchurch, New Zealand Taipei, Taiwan Bologna, Italy Choose two international hosts Santa Cruz, California Osaka, Japan
Results Measured Distance Matrix in milliseconds
Results Vector/Edge distances for Landmarks in two dimensions
Results Vector/Edge distances for Landmarks in six dimensions
Results Vector/Edge distances for Hosts in two dimensions
Results Vector/Edge Distances for Hosts in six dimensions
Effective in higher dimensions Conclusions Effective in higher dimensions Specifically six dimensions Host predictions become inconsistent Number of Landmarks? Worth investigating for future work Algorithm needs to be decentralized