6to4 borkeness (Was: Google and IPv6)

[Steve Wilcox]

On Thu, Mar 20, 2008 at 8:05 PM, Joe Abley <jabley at ca.afilias.info> wrote:

>
> On 20 Mar 2008, at 15:59 , Steve Wilcox wrote:
>
> > What is the advantage to take a v4 client accessing a v4 service and
> > then go and drop an asymmetric mtu reduced v6 tunnel in the middle
> > anyway?
>
> That's a good question for www.google.com.



> Another question is: if you assume that v4 clients are going to try
> and kill the user's experience by tunnelling requests over v6 before
> falling back to v4, is it better for that fallback to happen quickly,
> or not at all?


If you have a choice between "always works" and "may randomly failure in
weird ways" .. in the real world the answer is obvious...

Neither of these questions are especially pertinent for
> ipv6.google.com, though, since (AIUI) you can't reach that over v4 (at
> least, the DNS gives no clues as to how to do so).


ipv6 has only AAAA records and is imho probably more useful that way for
now, there are various mixed v4/v6 sites out there tho if you want to test
anything..

However if either www or ipv6 were to have both A and AAAA then we get back
to the matter in hand that is which protocol to use, and what decision to
make on selecting protocol, failing etc..

Steve

-- 
Global Infrastructure
Google Inc.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.cluenet.de/pipermail/ipv6-ops/attachments/20080320/94362cc0/attachment.htm