On killing IPv6 transition mechanisms

Gert Doering gert at space.net
Tue Mar 16 15:52:47 CET 2010


On Tue, Mar 16, 2010 at 12:01:38PM +0000, Benedikt Stockebrand wrote:
> Gert Doering <gert at space.net> writes:
> > I keep wondering about those 150ms.  I seem to remember that Google saw
> > *few users* that had a much higher IPv6 latency - but at the same time
> > they also saw users with a *lower* latency via IPv6 (due to different
> > network paths being taken).
> As I understand it they pick who they actually advertise AAAA's to, so
> their numbers are statistically biased.  What would be interesting to
> know was what the latency looked like if they provided AAAA's to
> everybody.

As far as I understand, these numbers are not for the "google over IPv6"
service, but for a certain percentage of the "standard" search requests 
where they serve v4-only, v4+v6- and v6-only pixels on the HTML page, and 
then compare the resulting requests for these pixels.  Or something like 

> > For me, the v4/v6 latency is very similar - I use v6 all day, many of
> > the sites I connect to (including google) have v6, and I do not see any
> > adverse effects (nor any positive effects - it's just IP, after all).
> Do you get the same result when you connect to a statistically
> significant selection of IPv4-only ISPs using whatever 6to4 public
> relay?

I'm not exactly sure how a 6to4 public relay would enable me to connect
to an IPv4-only ISP?  (And I don't have a high enough number of servers
on 2002:: addresses to make this a meaningful experiment).

If I connect to an IPv4-only ISP, I use IPv4 on my end...

> On another track: Much of the discussion in this thread makes an
> implicit assumption that there's a single DNS name for both IPv4 and
> IPv6 services, i.e. "www.example.com." has both an A and a AAAA
> record.  

Which is the ONLY thing that makes sense in the long run.

> That's fine if have the resources to do the Google approach
> and only serve AAAA's to whitelisted clients.  If you can't do that,
> then there's the alternative approach of using "www6.example.com." for
> AAAA's---that requires conscious action on the client side.

Which was OK for testing IPv6 services in the last century, but this
is completely useless as a way forward.  Why should anybody (except
somebody wanting to test this) type "www6" or "www.ipv6" or "www.six" 
into their web browser?

> These points aren't only crucial to keep in mind if you do business
> with end users; they also apply to end user acceptance of IPv6 and as
> such the move towards IPv6.

There is no "end user acceptance of IPv6" - as the end users do not know
what IPv4 or IPv6 is, they will use it if it is there, and not use it 
if it is not there.  Which is why the argument "my customes are not
asking for IPv6" is not very useful for this sort of customers either.

The challenge for the large-scale providers is to roll out IPv6 to the
end customers in a transparent way and without breaking anything in the
process - and do this before the NAT4444 architecture is so cemented
in all the heads that we're stuck with a completely no-more-end-to-end
Internet for ever.

Gert Doering
        -- NetMaster
Total number of prefixes smaller than registry allocations:  150584

SpaceNet AG                        Vorstand: Sebastian v. Bomhard
Joseph-Dollinger-Bogen 14          Aufsichtsratsvors.: A. Grundner-Culemann
D-80807 Muenchen                   HRB: 136055 (AG Muenchen)
Tel: +49 (89) 32356-444            USt-IdNr.: DE813185279

More information about the ipv6-ops mailing list