Re: IPv6 traffic stats - limitations of 6to4

[Pekka Savola]

On Thu, 13 Nov 2008, Rémi Després wrote:

 If an implementation implements RFC3484 and the user is not using custom
 address selection policy, all compliant RFC3484 implementations should
 prefer v4 when connecting to native from 6to4 (rule 5 of destination
 address selection AFAIR).


Actually, my above statement is incomplete.  Thanks for your eagle 
eyes :-)


In case the user has a RFC1918 IPv4 address and the destination is 
global v4 address, you'd use 6to4.  In case IPv4 address is global and 
destination is global, or both are RFC1918, you would use IPv4.


As such:

Can we derive from this that Google's IPv6 address is necessarily 
6to4 (most of its US customers using it are 6to4), and that Google 
has therefore a guaranteed path toward other 6to4 hosts?


I believe Google is using native addresses.  The 6to4 hits are 
probably caused by the users with private v4 addresses or users whose 
implementation does not support rfc3484.


Besides, isn't this a strong reason in favor of native IPv6 (albeit like Free 
did it with 6rd on its IPv4 infrastructure) vs 6to4?


Native is in many cases better than 6to4 or Teredo (but in some cases 
6to4 - 6to4 is better than native).  But this is something I 
specifically didn't comment on in my mail.


--
Pekka Savola You each name yourselves king, yet the
Netcore Oykingdom bleeds.
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings___
Ietf mailing list
Ietf@ietf.org
https://www.ietf.org/mailman/listinfo/ietf

Re: IPv6 traffic stats - limitations of 6to4

[Rémi Després]

Pekka Savola   (1-12/1-31/200x) 11/12/08 9:09 PM:
If an implementation implements RFC3484 and the user is not using 
custom address selection policy, all compliant RFC3484 implementations 
should prefer v4 when connecting to native from 6to4 (rule 5 of 
destination address selection AFAIR).
Can we derive from this that Google's IPv6 address is necessarily 6to4 
(most of its US customers using it are 6to4), and that Google has 
therefore a guaranteed path toward other 6to4 hosts?


Besides, isn't this a strong reason in favor of native IPv6 (albeit like 
Free did it with 6rd on its IPv4 infrastructure) vs 6to4?


RD


FWIW, in Linux this was changed as the default maybe about 2-3 years 
ago.  I suppose may other operating systems, especially recent ones, 
also operate in this manner. For Linux, some info is here: 
http://people.redhat.com/drepper/linux-rfc3484.html


This has been discussed on v6ops and ipv6 lists but unfortunately I 
can't find the threads despite search attempts.


Maybe someone else with better memory could provide better references.

This is why observing ipv6 traffic on a dual-stack hostname will 
mostly just in observing those that use native v6 (with Mikael, this 
was 0.5% of users).
Except if the dual stack server, like that of Google, uses different 
URLs for IPv6and IPv4, right?
If you're interested in wider picture of IPv6 penetration, you'll put 
the content on v6-only hostname (with Mikael, this was reachable by 6% 
of users). If you want to also cover for Vista users with Teredo, 
you'd put the content on a site and refer to it using a numeric 
address instead of a hostname (this would result in even a higher 
percentage).


So, if you're interested in any kind of IPv6 connectivity at all (even 
6to4, teredo, ...), at least in some user communities (p2p users), I'm 
pretty sure IPv6 penetration is already over 10%.  At least 6% is 
already proven by measurements :-)




___
Ietf mailing list
Ietf@ietf.org
https://www.ietf.org/mailman/listinfo/ietf