Abstract: Performance metrics between any two arbitrary end-hosts are predicted based upon previous measurements on the Internet between a plurality of geographically dispersed vantage points and clusters of end-hosts. Each cluster comprises end-hosts that are related based upon their IP address prefixes. In response to a central agent that stores the measured data for each of a plurality of predicted paths on the Internet, the vantage points use traceroute software to measure and periodically update performance metrics such as latency, bottleneck capacity, bandwidth, and packet loss rate for links comprising the predicted paths between the vantage points and one (or more) destination points associated with each cluster, and gather such data using distributed application systems. A user or client application can subsequently request predicted performance metrics for communication between specific end-hosts, based upon the previous measurement data.