Connectivity to an IPv6-only site

Fri Apr 23 19:38:50 UTC 2010

----- Original Message -----
> From: "Leo Bicknell" <bicknell at ufp.org>
> To: "NANOG" <nanog at nanog.org>
> Sent: Saturday, 24 April, 2010 7:33:21 AM
> Subject: Re: Connectivity to an IPv6-only site

> In a message written on Fri, Apr 23, 2010 at 01:08:30PM -0400,
> Valdis.Kletnieks at vt.edu wrote:
> > No, the problems are probably further back in time. We first started
> > turning up
> > IPv6 back in 1997 or so. There's a *very* good chance that we turned
> > it off a
> > decade ago (or whenever people *first* started listing quad-A's in
> > NS entries)
> > due to breakage and never actually revisited it since then. This
> > would have
> > been in the era of early 6bone and "your IPv6 connection is probably
> > tromboned through Tokyo".
> 
> Back in that era there was a very real problem of islands. That
> is, a group would set up IPv6 internally but never connect to the
> "Internet" (however you want to define that). So they got a AAAA
> and blackholed trying to reach it.
> 
> When you look at the content providers (Yahoo and Google tend to
> speak about this) they are very concerned about this problem as end
> users can make themselves islands fairly easily (an island of your
> house, for instance).
> 
> While the numbers are troubling for them, they are actually really
> good news. Depending on who's number you believe and when somewhere
> between 0.01% and 0.5% of end users are on unconnected islands.
> Now, when you serve a billion page views a day, dropping 0.5% is a
> huge concern; but it actually means the island problem has gotten
> really small.
> 
> More importantly, those are end users who are islands. Someone
> who's airport is misconfigured making them appear to have IPv6 when
> they do not. Most of these folks don't run recursive name servers.
> While I don't know of any hard data, I would expect the number of
> nameservers in islands to be at least one, and perhaps two or three
> orders of magnitude less.
> 
> So, in the context of publishing AAAA's for your nameservers, I think
> things are extremely safe at this point. If the recursive box on the
> other end has IPv6 at all and tries to use the AAAA there is a very
> good chance it will have working IPv6.
> 
> In the context of publshing AAAA's for your services (e.g. WWW),
> you need to look at the Google and Yahoo stats network wide, look
> at your own user base, and determine what level of breakage is
> acceptable. Keep in mind that IPv4 doesn't always work, so 0% is
> an unachieveable goal. :)

Well google will not serve you an AAAA record if you are not registered with them. This to avoid all the issues above. Once you are registered, expect lot of IPv6 traffic!