Re: IPv6-only devices?

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

On Thu, 6 Feb 2003 21:56:43 +0100
Ronald van der Pol <Ronald.vanderPol@rvdp.org> wrote:

> On Thu, Feb 06, 2003 at 15:39:30 -0500, Keith Moore wrote:
> 
> > It's a lot simpler than trying to teach applications to
> > cope with a mixture of the two - which gets you back to the problems
> > inherent in scoped addresses.
> 
> I don't agree. Applications that walk all addresses (v6 and v4) are
> widely deployed without any problems. I am thinking about ssh, http,
> smtp, etc. Porting applications to use both v4 and v6 addresses is
> in many cases easy.

This works great for some applications; really poorly for others.
In general it works for apps that only involve two parties, and thus
two network endpoints.  There the rules are simple.  If both speak
v6, use v6.  Else, if both speak v4, use v4.  Otherwise, you may be
out of luck.  (actually this is likely to cause painful delays for
dual-stack hosts, but for the moment let's ignore that problem)

Surprisingly few apps work this way.  For instance the web doesn't work
this way.  The vast majority of the content on the web is accessible
only via IPv4.  If your client can't get to IPv4, you lose.  Email
doesn't work this way either.  Everyone needs an MX that speaks IPv4.
Yes, you can use a web proxy to allow your v6 client to get to the v4
web, and you can use a mail relay to allow your v6 client to relay mail
to the v4 mail network.  Both protocols already support a form of
proxying, so no protocol changes are necessary.  But you still end up
requiring an operator to make protocol-specific arrangements in order
to permit allow cross v4-v6 operation. And if you want your v6-only
web server or MTA to be accessible from v4, you're going to need to make
special arrangements to do that, also.

> It seems like what you are promoting is balkanizing the internet.
> Some nodes are v4-only and can only talk to v4-only. Other nodes
> are v6-only and can only talk to v6-only. Is that what you want?

It's unavoidable and inherent in the fact that there are two different
address realms.  Trying to mix v4 and v6 has the same problems that
exist when trying to mix SL and globals. 

What I'm pointing out is that trying to make every application speak
both IPv4 and IPv6 is a waste of energy. It's not even possible
in general, and for many applications there are going to be inherent
biases toward either v4 or v6 - e.g. for backward compatibility
reasons, because the application inherently requires a large flat global
address space, or because the application was initially deployed in
portions of the network (e.g. the cell phone newtork, or certain parts
of the world) that were more friendly to v6 than to v4.

Keith