--On Saturday, 26 April, 2003 13:02 -0400 Keith Moore
<moore(_at_)cs(_dot_)utk(_dot_)edu> wrote:
we need to get rid of site-locals. merely renaming them as
private use addresses wouldn't solve any of their problems.
there's no advantage to moving to IPv6 if it repeats the RFC
1918 mistake.
Keith,
Given the tone of some of the traffic on these subjects in the
last few weeks, this probably needs to be said in public, and I
apologize for singling you out because you are certainly not the
only offender.
Hyperbole does not serve either the community, or one's own
position, well in this situation or others like it. Suppose we
start making a list of problems that IPv6 doesn't solve
--routing, security, cancer, even NATs -- or just carries
forward from IPv4. If we then take that list and use it to
construct statements of the form of "there's no advantage to
moving to IPv6 if it doesn't solve X or repeats Y", we would
really be asking the community to forget the problems IPv6 does
solve (or at least helps with).
If we were using that approach and logic in the mid-70s, we
would probably be trying to run the network on NCP --and an
address space of 254 or so hosts-- today, presumably with either
the demon grandchildren of NATs or some offspring of a NAT and
an X.75 gateway.
Specifically...
* Like you, I'd like to get rid of NATs. I agree that
private address spaces tend to encourage NATs, although
not as much as running out of address space does. If
IPv6 is able to reduce the number of NATs by even
25%-50%, that would, IMO, create a significant advantage
for moving to IPv6. Getting rid of all of them would be
better, of course, but claiming or implying "no
advantage" if all of them can't be removed is really a
bit much, don't you think?
* Like it or not, there seems to be a strongly-perceived
requirement for non-RIR-dependent (neither PI nor PD-but-
RIR-allocated) addresses. That requirement is arguably
legitimate for completely detached networks, even if one
believes in globally-unique addresses for every network
or host that is even vaguely connected. Many years ago,
I could come to the NIC and say "I need some
(globally-unique) addresses for a LAN that will never
be connected to the Internet" and get them. Today,
the RIRs have no procedures for dealing with such a
request, much less making the allocation, in either IPv4
or IPv6 space. I would hope that, were someone to
make it, he or she would get an explanation, rather than
polite laughter, but the requirement is there. And it
seems to me that we either need to figure out how people
can get addresses for that narrow purpose or reserve
some "open-season" addresses for them. That is clearly
independent of whether it is rational to have SL
addresses on hosts that might be connected to the public
network --either through NATs or multiple addresses--
but it means that having some such addresses around
doesn't turn IPv6 into "no advantage".
* There has been increasing evidence, as this set of
discussions have played out, that SL isn't the problem,
but merely a symptom of a series of issues with scopes
and/or routing and/or layering and data abstraction. If
we figure out how to solve the appropriate set of those
issues, SL will either go away or not, but it is
unlikely to be a big issue. Absent a huge intellectual
breakthrough (and a model for deploying it easily),
these look likely to work out as complex tradeoffs in
which we need to make decisions about optimality, or
damage-minimization, rather than coming up with a
perfect solution that meets all needs and creates
universal pleasure. Are you going to pick some point in
that solution space, or the subspace that doesn't happen
to contain SL, and say "if we don't get to this point,
IPv6 is worthless"? I don't think so or, at least, I
hope not.
So, if possible, let's try to calm it down a bit, stop the
hyperbole, and get back to thinking and reasoning carefully
about a complex problem. Please.
john