Systematic Reconstruction of the Past , Present , and Future of Internet Path and Delay DK

Delay and path between arbitrary points in the Internet are not always measurable and often hard to estimate or predict. In this work we use the path stitching algorithm to systematically reconstruct the past, present, and future of Internet path and delay. The input to path stitching is the traceroute output data from CAIDA’s Skitter and Ark projects and DIMES and routing tables from RIPE and RouteViews. Without any additional active measurement, path stitching can produce delay and path estimates between random host pairs. Comparison of the estimated delay with PingER data demonstrates that the estimates are close enough to live measurements. The six years of the reconstructed history of the delay distribution between random pairs of /24 prefixes demonstrates how the overall Internet delay performance has evolved over time. Path stitching relies on AS-level path inference. By changing the AS path we can answer what-if scenarios of AS path change on end-to-end delay performance. Our results demonstrate that path stitching is a powerful and versatile tool for a wide range of applications. It also opens up a new direction in performance estimation that takes full advantage of structural properties of the Internet.