[158115] in North American Network Operators' Group
Re: Big day for IPv6 - 1% native penetration
daemon@ATHENA.MIT.EDU (Geoff Huston)
Wed Nov 21 07:14:18 2012
From: Geoff Huston <gih@apnic.net>
In-Reply-To: <F8230BA6-D4DA-4621-9417-E6BC052A5E92@ianai.net>
Date: Wed, 21 Nov 2012 23:12:46 +1100
To: Patrick W. Gilmore <patrick@ianai.net>
Cc: NANOG list <nanog@nanog.org>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On 21/11/2012, at 3:05 AM, Patrick W. Gilmore <patrick@ianai.net> wrote:
> On Nov 20, 2012, at 08:45 , Owen DeLong <owen@delong.com> wrote:
>=20
>> It is entirely possible that Google's numbers are artificially low =
for a number
>> of reasons.
>=20
> AMS-IX publishes stats too:
> <https://stats.ams-ix.net/sflow/>
>=20
> This is probably a better view of overall percentage on the Internet =
than a specific company's content. It shows order of 0.5%.
>=20
> Why do you think Google's numbers are lower than the real total?
It depends on what you are trying to measure and how you are measuring =
it.
I don't know Google's methodology, but lets say its a simple form of the =
experiment:
"When presented with a dual stack object what percentage of users prefer =
to retrieve that object using IPv6 as compared to IPv4?"
Up so a year or so ago if a browser had access to IPv6 and IPv4 it would =
first attempt to connect using IPv6 and if the connection failed then it =
timed out and then tried to use IPv4. So the experiment would be roughly =
commensurate with measuring working IPv6 systems on end sites connected =
to workin ipv6 access networks of one sort or another.
More recently some browsers (Safari on Mac OSX, Chrome, Firefox with =
config settings enabled) have adopted a different strategy and when =
presented with a dual stack object some clients may end up trying the =
connection using IPv4 first and then fall back to IPv6 if IPv4 fails or =
times out. If the experiment simply counts the percent of clients who =
prefer to connect using IPv6 in a Dual Stack scenario, then some of =
these users running more recent versions of the browser will not be =
counted.
There are ways to compensate for this, including running a series of =
tests, and this form of approach is described at =
http://labs.apnic.net/measureipv6/
I personally have no knowledge if the numbers published by Google =
reflect the "prefers to use IPv6 in dual stack mode" or "is capable of =
using IPv6 (by virtue of being able to retrieve a IPv6 only object)" =
These days the second number is larger than the first.
Geoff
