Thus spake "Robert Elz" <kre(_at_)munnari(_dot_)OZ(_dot_)AU>
| which is that a large part of the Internet is going
| to continue to be IPv4-only.
It simply cannot be. Either it dies (or atrophies, which is
essentially the same thing), or IPv4 is replaced by something.
It is possible that hosts with existing IPv4 connectivity will continue to
use it and never upgrade to IPv6 -- only new hosts would use the latter.
With NAT, it's possible that many end sites (even new ones) may use IPv4
internally and NAT to IPv6 for external communication.
IPv4 is dead. Long live IPv4.
No matter what we do, there is not enough IPv4 address space for
the core network to use to reach future end sites.
Arguably there is not enough IPv4 address space to meet current needs, but
we're hiding that problem with the use of NATs. Some places are so starved
for addresses they're even using multiple layers of NATs. Good luck
figuring out how many IP-connected hosts are really out there now; I doubt
it's up to a billion yet, but it'll probably be there soon (and 25%
efficiency is the norm for IPv4).
The future is obviously bad for IPv4, as it doesn't even have enough
addresses for each human on the planet, much less for the dozens of
"connected" devices per person that have been theorized.
And, of course, those are absurd assumptions (flat routing /28 indeed!)
Why mess with /28s? If we're going to mess up routing, we might as well
assign /32s to each host and route those. Nope, still not enough addresses.
The core network, the part that needs to be able to identify end
sites, simply needs more addresses - and there's no way to fake it
via layers of NAT, end nodes from any random point on the net still
need to be able to indicate that it is my site (forget even about host),
and not yours, that they want to contact. That means that we have
to have different addresses (identifiers). IPv4 simply cannot last.
That's why several years ago we moved from just the IP identifying the host
to using port numbers as well (aka NAT). 2^48 hosts is doable, though
messy. Oh, and that doesn't include the hosts that aren't talking
externally and thus don't use any address+port resources at all...
Of course, end user sites could keep using IPv4, and use something
like NAT to translate, to whatever the code network is using. And
for a while, no doubt some will continue. But why would that do
that indefinitely, it makes no sense? The translation is just an
extra cost/management burden that they don't really need - after all,
all (at least if it is IPv6 that the core switches to), all the
end-user equipment that counts already supports IPv6, switching end
sites now is easy (and relatively painless).
The cost of IPv4 NAT is apparently perceived to be lower than the cost of
IPv6 and 6to4.
| So, what's the functional difference between:
|
| - A host which has an IPv6 only address, which it cannot use (without
| "borrowing" a global IPv4 address) to comunicate directly with
IPv4-only
| hosts out on the global Internet.
|
| - A host which has an IPv4 local-only address, which it cannot use
(without
| "borrowing" a global IPv4 address) to comunicate directly with other
IPv4
| hosts out on the global Internet.
short term, of course, nothing. But the former has the possibility,
even expectation, of the "borrowing" simply ending, most likely even
in the not too distant future (though certainly beyond the end of the
financial year, and so way out of scope of many people's thoughts...)
The former does have an obvious path to a full IPv6 deployment, but (a)
there is a serious "first mover" problem, and (b) the costs of taking that
path will be paid twice.
Even now, with Windows finally supporting IPv6, it's still tough to find
SoHo/consumer FW devices that will pass IPv6 traffic (even tunneled) and
many corporations are forced with swapping out every L3 switch they own to
get models that support IPv6. There is substantial economic cost to the
former scenario over the latter today, even before you calculate the
training cost of millions of MCSEs and CCNAs that don't know anything about
IPv6 and see no reason to learn.
I predict things will continue roughly as they are now, and when the IPv4
space is approaching true exhaustion the prices of PI and PA space will rise
so much that it will exceed the cost of converting to IPv6. Then IPv6 will
take off, and not before.
S
Stephen Sprunk "God does not play dice." --Albert Einstein
CCIE #3723 "God is an inveterate gambler, and He throws the
K5SSS dice at every possible opportunity." --Stephen Hawking
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf