Re: dual stack & IPv6 on by default

[Tim Chown <tjc@ecs.soton.ac.uk>]

On Mon, Mar 10, 2003 at 06:04:02PM -0500, Bound, Jim wrote:
> 
> For edification.  I have a node on a work LAN that knows nothing of
> IPv6.  I download software and configure my node to be capable of IPv6.
> I manually configure my interface to support IPv6.  I now ftp to an IPv6
> address.  This is not going to work.  But I was stupid to think it
> would?  Is this the kind of basic mistake your also worried about?

Agreed if you're static in a network, then that's a mistake.  But a common
problem is taking an IPv6-enabled device from a "dual stack" network where 
it happily runs using either protocol into one where only IPv4 is supported.
Then what happens varies with OS/browser.   

For example, in WinXP+MSIE6 any attempt to reach a site with an AAAA (and A)
record will fail with an error after 20-60 seconds.  With WinXP+Mozilla 1.3 
a similar attempt will just get an immediate "connection refused".   In 
neither case is there (the desirable) fallback to IPv4.

So for people who want to go dual-stack in their workplace yet take devices 
to IPv4-only networks that's the problem they face.   Of course you get what 
you deserve maybe for the application writer trusting presence of a DNS 
record as an indication of connectivity.

There will be some user-land settings required for IPv6 connectivity.  For
example RFC3041 needs to be per application so you can rely on (as many 
people will want to) per-host IP-based authentication for ssh access, while
taking advantage of RFC3041 while browsing the net from the same host.
I can see that we can offer users a privacy/fixed IP toggle, and many could
understand that as an advanced setting in their TCP/IP options, but it's
something we want to shield "typical" users from.

I don't believe a userland v4/v6 toggle would be useful for "typical" users.
It be used by the v6-geeks in the current stage of deployment, but we need
more robust methods for handling movement between dual-stack and IPv4-only
networks where IPv6 is enabled on a device.   We're paying for the choice
of app developers to promote IPv6 by trying IPv6 ahead of IPv4, and then not 
falling back (quickly) when it fails.

Tim