Re: IPv6 transition architecture discussion

[Keith Moore <moore@cs.utk.edu>]

> Suggestions on how to deploy IPv6 from the operating 
> system perspective seem to be lacking. 6to4 looks like a good solution, 
> but it doesn't work behind NAT. In addition, people in this working 
> group have strongly advised against any wide deployment of 6to4 hosts.

I don't think there's consensus about that, only that there are some 
questions or problems associated with 6to4 on a host level that haven't
been fully resolved.  Personally I find "host 6to4" essential both to
provide access to machines at home (where a single host serves both
as an IPv4 presence and as a v6 router) and to allow me to access my 
v6 hosts from v4-only net connections.  I'm really disappointed that so
few host OSes support it.  For instance the lack of 6to4 support in 10.2
probably means I'll end up running NetBSD on my new 1Ghz powerbook g4.

> Apple would like to ship a version of our operating system that does 
> everything it can to get an IPv6 address. 

I'd love for MacOS to support 6to4, though I'm not sure that having
it automagically enabled is a good idea.  (there's such a thing as
a machine being too clever or having too high an opinion of its
own cleverness)

> The vocal people in this working group say that 6to4 and 
> Teredo are bad solutions. They rightly point out that the few 6to4 
> relays will probably melt down or people will stop running them. 

6to4 works wonderfully if you have v4 connectivity (perhaps indirectly)
and are talking to other 6to4 sites.  if you want to talk to other sites
over v4, right now you generally need to set up an explicit tunnel for
that.  the anycast based relay routers don't work well enough to be used,
IMHO.  at least not right now.  attempts to talk to v6 sites typically
time out (which is annoying because the v6 addresses are tried first -
another thing which needs to be fixed)

> We look to the IETF for a solution because we can't solve this problem 
> alone. I get the impression from this working group that the working 
> group is of the opinion that the transition mechanism will not work and 
> we should just sit on our thumbs until the ISPs provide us IPv6 
> connectivity.

be assured that not everyone here shares that opinion.  IMHO we need
to define a host profile for 6to4, and perhaps another one for NAT boxes
that wish to support 6to4.  yes there are problems.  let's fix them.

> We were considering turning on 6to4 by 
> default when there are no routing advertisements. This working group 
> has dissuaded us from doing so. If we did turn on 6to4, we would want 
> to add some extra smarts to getaddrinfo to list IPv4 addresses first 
> when 6to4 is in use to reduce the load on the relays. 

it turns out that you need the address ordering to be sensitive to both
network configuration and to the particular application.  some apps will
be v4 by default, others will be v6 default, others will be exclusively 
one or the other.

Keith