IPv6 statistics at tronche.com and related sites

What are these statistics and how are they collected ?

The goal is to measure IPv6 penetration, from the point of view of a web server. The servers are dual-stack, both IPv4 and IPv6 addresses are published in the DNS. It is expected that the client (OS + browser) favors IPv6 over IPv4 in these conditions.

3 indicators are collected: the number of hits (HTTP query), the number of distinct IP addresses, and the number of distinct IP networks, as advertised in whois databases.
Although collected, I consider the number of hits not being a good indicator for the following reasons:

  • Given the total hit numbers, which is not very high (~200K / month), it is sensitive to the some particular operations. For example, the surge in March is related to me fetching as many pages as possible from the server to "stress test" my IPv6 connection. If that kind of operation, from one individual, has a significant impact, and this is the case, it is too sensitive to statistic noise.
  • The ratio of IPv6 to IPv4 adresses if around 1%, more or less in line with recent studies (see here for an example, another one, in french). In contrast the IPv6 to IPv4 hits sometimes peak at 7%. This is not surprising, since this site is still one of the rare IPv6 enabled sites, people with IPv6 (should I say geek ?), tend to "play" with it and spend a long time on IPv6 enabled sites, just for the sake of it.

The number of addresses has its flaw too, since the IPv6 addresses is supposed to change from time to time to protect privacy, so it is a weak indicator of the number of distinct visitors (weaker than in the case of IPv4 ? I'm not sure).

The data

Hits

Distinct IP addresses

Zoom on IPv6, splitted by RIR

Hits

Distinct IPv6 addresses

Distinct IPv6 networks