Re: [v6ops] “ipv6 on windows 7” - Google Search

[Mark Smith <[e-mail]>]

Hi Fred,

On Wed, 6 Oct 2010 09:14:28 -0400
Fred Baker <[e-mail]> wrote:

> > It would be interesting to understand what the issue is here. I suspect it has nothing to do with the actual Windows (or Apple, Linux, etc) implementation, but is more related to issues as discussed in http://tools.ietf.org/html/draft-wing-http-new-tech. Would it make sense for v6ops to make a recommendation to address the concerns?
> 
> Reading over my own email, my comment can be confusing - the getaddrinfo() stuff is in the implementations in question, so it does indeed "have something to do with the implementation". I guess my question is - 
> 
> I get questions about operational IPv6 deployment due to issues of implementation. One ISP recently told me that they can't deploy AAAA records for certain services because when their customers are in the region of a neighboring ISP that runs IPv6 as a walled garden and btw respond to connection attempts with a TCP RST, their user's laptops don't correctly fall back to IPv4. I can imagine several possible recommendations, and might actually recommend two of them: "happy eyeballs" as Dan Wing describes (open TCP sessions at some rate trying all IPv4 and IPv6 combinations until one of them opens a session, with the objective of achieving an open session within a stated interval such as 1 second or less), and an operational recommendation that if you're going to run IPv6 as a walled garden, that you configure your first hop routers to respond "host unreachable" or "network unreachable" to any packet that you have decided to not deliver.
 
A while back I converted an application to use getaddrinfo(), and
the advice in the manual page, when there were multiple A/AAAA
responses, was to walk through them all, in order and in turn,
attempting to connect(). If connect() failed, regardless of the
cause e.g. timeout or connect refused, move on to the next result of
the getaddrinfo() call.

Some advice in regards to how getaddrinfo() should be used might be
useful, although I wonder if it would be better placed in an update to
one of the IPv6 socket API RFCs. Some text relating to the use of
getaddrinfo() (an an appendix possibly) in an update to the source
address selection RFC probably wouldn't hurt though.

I can't really see why this 'walled garden' ISP is responding with TCP
resets though. I'd have thought it would be more customer friendly
(more "garden like") to send an connection attempts to
unknown/unavailable destinations towards something like a wildcarding
web server.

Regards,
Mark.