We have major issues geolocating ipv6 traffic, caused by either data missing from the GeoIP database, and/or by a bug in the CentralNotice geoip code.
When validating the ipv6->v4 fallback code, let's find a way to measure how many people * begin the fallback code * succeed in falling back to ipv4 geoip.
From @Ejegg:
For the hour we've been examining, the US had 1.36M en.m.wikipedia.org page views and 1.17M impressions on v4 networks (86% impression rate). On v6 networks, it had 387k page views vs only 27k impressions (7% impression rate). Other countries had comparable discrepancies, though their v6 networks are a smaller proportion of their total traffic.
US-mobile will be hit the hardest, these are real measurements of the proportion of traffic coming in via ipv6:
15:32 < paravoid> 6.5% were to en.wiki, 13.4% to en.m.wiki