Factors, actions influencing the possibility/timing of IDR for IPv6-basedrouting domains?

Anthony Roberts ipv6-ops at arbitraryconstant.com
Wed Jun 24 07:17:23 CEST 2009


On Fri, 19 Jun 2009 18:32:25 -0700, Erik Kline <ek at google.com> wrote:
> One thing we try to consider is that the only thing you can assume
> about the IP connectivity of a user making a request is the protocol
> over which the request was made.  In other words, if a request comes
> in over IPv4 you can assume the user effectively has v4 access, and
> similarly for v6.  Right now you can be *statistically* assured that a
> user making a v6 request has v4, but that is not 100% true now, nor
> will it always remain even a strong likelihood.  (Of course, in
> theory, you can correlate several out-of-band factors to increase the
> accuracy of any guess about non-request-protocol availability, but why
> go to all that work.)  So if a request comes in over v6 we serve v6
> cache links in the response (bugs not withstanding :).
> 
> It's theoretically possible that if the IPv4 web corpus and the IPv6
> web corpus become significantly divergent that some accommodation
> might be made.  It might be more akin to search results in a language
> that wasn't your detected native language, though.  Too early to tell.
> 
> On a different note: one thing that will be nice when v6 becomes
> popular with hosting providers: everyone can get a static v6 address
> and hence an SSL cert (i.e. we won't need SNI).

It's a bit of a catch-22. Websites aren't going to go AAAA-only until the
clients lost by doing so are negligible, and IPv6 clients aren't going to
give up v4 connectivity until the utility of v4 is negligible. If true,
that means v4 will be the lowest common denominator until v6 is common
enough to take over. In turn, that implies v6-only content will only be
significant after a great deal of the migration is complete. Mitigation
like SNI will be needed long before then. When things get bad enough, I bet
we'll even see VPS/colo providers offering v4 reachable layer 7 proxies for
customers who can't afford public v4 IPs.

The thing I find encouraging are the NAT64 drafts, as they explicitly
acknowledge and address the fact that v4 may be with us for some time to
come. That may cause a problem for Google cache links though; you'll see v4
connections, and send back links with v4 IPs, but the clients will be
relying on DNS translation that won't happen for web links with raw IPs.
Good times. :)

Regards,

-Anthony


More information about the ipv6-ops mailing list