On 26-04-2003 19:35PM, "Keith Moore" <moore(_at_)cs(_dot_)utk(_dot_)edu> wrote:
What is wrong with having addresses available for private use on
networks that do not intend on being connected to the Internet?
in principle, nothing. but experience has shown that most of those networks
do end up being connected to the Internet, while still keeping those addreses,
and that applications are expected to cope with that.
Ehm... What experience? You are referring to experiences with RFC1918
addresses in the IPv4-world, aren't you? But... Do you or anybody already
have have similar experiences with regards to site-locals?
Indeed, RFC1918 addresses are frequently used behind NAT-boxes. And these
NAT-boxes do create all kinds of troubles. That is the main problem isn't
it?
For that matter will it be the same with regards to IPv6 site-locals? I
think not. Because the most important reason for NAT is the lack of
available IPv4 address space. I know there are more reasons why people use
NAT. But this is by far the most important one. Because for many many
end-users and small companies, it is very hard, if not impossible or at
least expensive to get more than one IPv4 address.
This is not the case in IPv6. ISPs typically provide their customers with a
/48 of globally unique address space... Well, my home-ISP does. Do you
really think I thought one minute of NATing IPv6?
In other words: NAT makes networking complex, it breaks the end-to-end
model, creates unwanted site-effects and it is not really needed with enough
available address space. Therefore I do not think that site-locals will be
used in the same fashion as RFC1918 addresses are mostly used today.
Having said that, NAT might be a quick and dirty hack to satisfy the needs
for:
- address stability, due to the lack of easy renumbering;
- multihoming;
- (obscurity).
Especially renumbering-issues might also be a reason for using NAT nowadays.
Just because renumbering is too much pain in IPv4. Somehow addresses have to
be configured at many different places. Implementers might be more
thoughtful with regards to IPv6.
Ironically, multihoming is typically a requirement of non-NAT users
nowadays. They might become the new NAT-adepts if the market comes with
means to do it via NAT and there is no good alternative.
Obscurity (some seem to call it security) is just an implementation matter
as well. Really don't want to discuss this.
In any case, I do not think that deprecating site-locals will take away the
NAT-threat. Neither does it bring us closer to alternatives. At least
site-locals are easily recognisable so coping with becomes a possibility.
Arien