Confidence-Weighted Marginal Utility Analyses for Improved Internet Mapping

Many techniques have been proposed for discerning internet topology, but not so many of them have been studied for their utility: the returns in information that they provide as their number of measurements increases. We define a probabilistic model for representing a map built on uncertain observations. We propose new utility metrics and reuse existing metrics in combination with this model to examine the marginal utility of observations for maps made with three different tools: Skitter, Scriptroute, and Rocketfuel. We define classes of observations and quantify their relative utilities and discuss how these results can influence future mapping techniques.