Hypothesis-Based Comparison of IPv6 and IPv4 Path Distances

Abstract

Short end-to-end path lengths and faster round-trip times (RTTs) are important for good client performance. While prior measurement studies related to IPv6 primarily focus on various adoption aspects, much less work have focused on performance metrics such as these. In this paper, we compare the relative end-to-end path distances and RTTs when using IPv6 and IPv4 between PlanetLab nodes in Europe and different subsets of popular domains. In addition to providing access to multiple measurement nodes, the use of PlanetLab also provides a use-case driven report of running IPv6 experiments on this previously prosperous experimental platform for academic research. In particular, the study provides a first report on performing IPv6 experiments on PlanetLab, highlights the lack of IP support among PlanetLab nodes and limitations of state-of-the-art traceroute tools used for IPv6 measurements, and provides a statistical methodology that uses hypothesis testing to derive insights while accounting for such testbed and traceroute shortcomings. Our performance analysis shows (among other things) that the relative RTTs of the IPv6 paths are currently faster than the corresponding IPv4 paths, and that the fraction of pairings for which this is the case is quickly increasing across a wide range of domain popularities and domain categories. These findings suggest that there is incentive to use IPv6, which may impact the rate of further IPv6 deployment.