IPv6 measurement studies have focussed on measuring IPv6 adoption. However, there have been few studies on measuring IPv6 performance. Using a 4-year long dataset, we measure IPv6 performance towards dual-stacked websites. We show that latency to popular websites over IPv6 has improved over the years. 40% of ALEXA 10K websites are faster over IPv6 today with 94% of the rest being at most 1 ms slower. We also identify glitches in web content delivery that once fixed can help improve user experience in IPv6-only networks. We show that 40% of ALEXA 1M websites with AAAA entries used to (2009) exhibit complete failure over IPv6. These failures have reduced to 3% today. However, 27% of popular websites with AAAA entries still exhibit partial failure over IPv6. These partial failures are affected by DNS resolution error on images, javascript and CSS content. 12% of these websites have more than half of the content that belongs to same-origin source and fails over IPv6, while analytics and third-party advertisements contribute to failure from cross-origin sources. We further identify areas of improvements within the IETF\@. We witness that 99% of ALEXA 10K websites prefer IPv6 connections more than 98% of the time. HE tends to prefer slower IPv6 connections in 90\% of the cases. We recommend lowering the HE timer value to 150 ms which gives a margin performance benefit of 10\% while retaining same preference levels over IPv6.
Slides: ietf.org/proceedings/99/slides/slides-99-v6ops-sessa-a-longitudinal-view-of-dual-stacked-websites-failures-latency-and-happy-eyeballs-00.pdf
