Re: on NAT-PT

[Alain Durand <Alain.Durand@Sun.COM>]

Jun-ichiro itojun Hagino wrote:

> > Is there any paper documenting this 'operational practice'
> > and why you consider it a 'bad thing' ?
> > I would like to understand better why you claim that it is 'universally'
> > a bad thing and not just 'a bad thing' within your local environment.
> >    
> 	unfortunately there's no documentation.
> 	this is universally bad because
> 	(1) it requires every node to configure the translation prefix manually
> 	(kills spirit of autoconfigured hosts), or to cook up DHCPv6 option for
> 	it (horrid idea),
It could be a well know prefix. You won't have to configure a special 
prefix for that.

> 	(2) NAT or NAT-PT has to be invisible to the client, instead of
> 	explicitly visible (otherwise, why bother rewriting protocol contents?),
> 	and
I do no think so, au contraire. This is a last resort to enable 
IPv6-only apps
on Ipv6-only node to communicate with their Ipv4 only peers.
Requiring such nodes (or apps?) to have special knowledge that something
is needed to "bridge" to the other world does not seems unreasonable to me.
This is what we mean by "IPv6 nodes will have to pay the burden of the 
transition".

> 	(3) the idea where client is aware of things that happen in the middle
> 	is exactly the same as RSIP.
You'll be aware that things will happen. You will not have
to negotiate it, and that makes a big difference.

> 	our changelog shows that the feature (to specify translation prefix
> 	via environment variable, which in turn controls getaddrinfo(3)) was
> 	dropped in January 2000.
You can always put it back in ;-) Remember, this is for IPv6-only nodes,
the kind that is not yet largely deployed.

   - Alain.