Dual stack IPv6 for IPv4 depletion

Matthew Kaufman matthew at matthew.at
Fri Jul 10 06:42:57 UTC 2015


On 7/9/2015 6:31 PM, John Curran wrote:
> On Jul 9, 2015, at 9:02 PM, Matthew Kaufman <matthew at matthew.at 
> <mailto:matthew at matthew.at>> wrote:
>>> On Jul 9, 2015, at 4:07 PM, Owen DeLong <owen at delong.com 
>>> <mailto:owen at delong.com>> wrote:
>>> ...
>>> You are correct… In order for 20% of Google’s traffic to come from 
>>> IPv6 connected devices, there would generally need to be more than 
>>> 20% of all devices connected over IPv6.
>>
>> That doesn't follow at all.
>>
>> One guy who has v6 and really loves youtube can account for most of it.
>
> Matthew -
>
> That would be the case if the measurements of “IPv6 users” were based 
> on traffic or packet
> counts, but Google’s measurements are based on specific pairs of HTTP 
> connection attempts
> (one IPv4, and one IPv6) and the ratio of those which are IPv6 
> capable.  The measurement
> methodology is documented in the Google research paper -
> <http://research.google.com/pubs/pub36240.html>

Still can be accounted for with *fewer* than "20% of all devices 
connected over IPv6" (the opposite of Owen's claim). Possibly even far 
fewer, if many "devices" don't bother to visit Google via HTTP.

I do find it interesting that Google (and other's) graphs show much 
higher IPv6 penetration on weekends - I assume that's because 
ISP-provided CPE + default OS configs get you better chances of IPv6 
than you get using your IT department's machine image plus network 
infrastructure. Anecdotally: I have yet to work regularly at a facility 
that has IPv6 connectivity to the outside world from the WiFi networks 
that serve employee laptops. (Though for several years in the late 2000s 
I did get IPv6 addresses via RA and routing between floors (but not 
beyond)).

Matthew Kaufman




More information about the NANOG mailing list