On killing IPv6 transition mechanisms
tjc at ecs.soton.ac.uk
Tue Mar 16 14:54:00 CET 2010
On Mon, Mar 15, 2010 at 06:58:34PM +0100, Ole Troan wrote:
> see: https://sites.google.com/site/ipv6implementors/conference2009/agenda/10_Lees_Google_IPv6_User_Measurement.pdf?attredirects=0
> which shows that a dual stack host is about 150ms slower using IPv6 than IPv4, and that ~0.08% of the hosts have broken IPv6. (please take notice in the presentation that there is quite a bit of statistical uncertainty because of few IPv6 clients).
The key missing stat there is the difference in latency for native v6
clients that are dual-stacked. Since 66%(!) of the Google 'experiment'
is 6to4 users, and the dual stack is thus resuambly IPv4 + 6to4, then
the average 150ms delay difference isn't fair/representative. More of
a reason to just throw 6to4 away.
Interestingly free.fr is considered native although it uses 6rd tunneling,
but I guess it counts as native as the tunneling is within free.fr :)
Slide 18 shows MacOSX seems to be the big 6to4 offender, or rather the
Airport Extreme is... and the slideset suggests ISPs deploy more 6to4 relays,
not less use of 6to4. Is 6to4 enabled by default on the Airport Extreme?
More information about the ipv6-ops