A question came up in Fundraising yesterday that I now have results for --
basically it's how many page requests view the site via IPv6; and how much
can we trust ip geolocation on those IPs.
tldr + a little extra: We can trust IPv6 geolocation for rough pinpointing
of countries. Based on complaints we get from donors, that for those
countries with not great connectivity we should take the GeoIP resolution
with a large grain of salt. City resolution for the US seems to work OK
with the exception of mobile -- I have no anecdotal evidence for anywhere
else.
--
[snip]
Basically it comes down like this:
Last month we served 6,166,994,300 banner requests (served to every page in
a content namespace -- so not Special pages but pretty much everything
else.)
- of which 28,739,500 were IPv6 (0.47%)
- of which 39,800 failed to resolve to a country, even with IPv4 fallback
(0.0006%)
-- (in the noise floor we had 300 people resolve to anonymous proxy or
'European Union' IPs; there were no satellite provider or 'Asia Pacific'
resolutions recorded)
I had Ottomata do a filter on
geoiplookup.wikimedia.org (our IPv4 only
fallback host). In the period between 22:58 and 23:17 we served 344
requests. In that time period we had roughly 210,000 IPv6 requests. Meaning
that our fallback is important for ~0.16% of IPv6 users.
I still think we should, in fundraising, do an interstitial page because
that's a friendly thing to do. But for sitewide operations; I think the
success rate is fairly good. Most of the 'crummyness' that Josh routinely
reports is fuzzyness around borders -- e.g. people in India resolving to
being in Pakistan.
[snip]
~Matt Walker
Wikimedia Foundation
Fundraising Technology Team