[ppml] those pesky users...
owen at delong.com
Tue Mar 27 16:34:46 EDT 2007
> Lee, ARIN and the other RIR's need to admit a huge mistake was made
> over the IPv6 allocations and then go forward with correcting it.
ARIN and the RIRs are not the ones who made this mistake. This
was originally in the plan and the IETF scrapped it from IPv6 for
I still don't understand (some alleged privacy concerns).
Frankly, I still think that we should have a protocol where there is
an IPv6 /96 reserved (this isn't really a big deal, is it?) that
network to attach to the v6 world via a bilingual gateway which has
the ability to support v6-only, dual-stack, and v4-only interfaces.
That way, all v4 addresses could simply be encapsulated into
that /96 prefix in the v6 world (yes, the v6 backbone would be carrying
the entire v4 routing table as specifics of that /96, so what... They
have to still carry it as a separate v4 table the way it is today).
This would mean that v4 hosts would still have the same ability
to reach v4 hosts that they have today, and, v6 to v4 connectivity
would be nearly transparent for the v6 users.
> What you should have done was for ALL ipv4 assignments you should have
> AUTOMATICALLY made an IPv6 assignment of a number block. This
> would have
> eliminated the silly "fee for IPv4" and "fee for IPv6" different rate
> schedules. There would only be one fee for "IP addressing" that would
> never go away and never change whether you were using IPv4 or IPv6
> or both.
> It would have allowed people that wanted to experiment to just look
> up their
> IPv6 allocation and start announcing them, without the bother of
> a number authority and going through an allocation scheme.
Short of a reserved v6 prefix which maps to v4 numbering, this doesn't
make sense to me. If we're going to go to the trouble of maintaining
registrations, then, not everyone who has v4 needs v6 and there's no
reason for the RIRs to provide free registration services on this basis.
> For example when the telephone companies switched from 7 digit to 10
> digit dialing in this area, they just told everyone to start
> dialing the
> area code. So you immediately knew what both your 7 digit and your 10
> digit telephone number was.
The difference is that all that changed was how you dialed. You
a new 10 digit telephone number, you always had a 10 digit telephone
but you used to be able to get away with only specifying the last 7
While I agree with you that IETF should have done something like this
I don't think the RIRs have any control over this issue.
> What the RIRs did was basically equivalent to the telephone company
> announcing that due to the need to go to 10 digit dialing, they were
> going to change every telephone number in the book to something
> completely unlike what the prior number was.
Again, you really need to learn where the distinction is drawn between
what the RIRs do (register associations between numbers and
and assure uniqueness of such registrations) and what the IETF does
(architctural decisions and protocol development).
The IETF made the decision you criticize in your previous paragraph.
The RIRs are just along for the ride.
> The best thing going forward would be for ARIN and the other RIR's to
> drop the IPv4 and IPv6 fee schedules (the wavier is a joke anyway,
> is the point of an IPv6 fee schedule with a fee of $0) and replace it
> with a single IP allocation fee schedule that applies to both kinds of
> numbering, then for all current IPv4 holders that the numbering
> have assigned numbering for, just go ahead and assign IPv6 allocations
> at a 1 to 1 ratio. (for every single IPv4 address you get an IPv6
> It's not like there's any shortage of IPv6.
That's not feasible with the IPv6 architecture. To take this a bit
minimum allocation unit in IPv6 is a /64 network. The RIR minimum is a
/48 with very few exceptions. A /64 is approximately 4 billion times
number of IP addresses in the entire IPv4 internet, so, a 1:1 assignment
ratio simply isn't possible even at the smallest possible network block.
A single /64 would be more IPv6 addresses than ANY IPv4 holder
currently has. However, many IPv4 holders really need the ability
to create multiple distinct networks. IPv6 (for better or worse, and,
this is the IETF and not the RIRs decision) has returned to classful
addressing, at least to some extent. The new format has been set
up as NNNN:NNNN:OOOO:SSSS:HHHH:HHHH:HHHH:HHHH:HHHH
where NNNN:NNNN is a 32 bit network number, mostly intended
to represent the LIR. OOOO is 16 bits which would generally
represent which Organization the LIR assigned the subordinate
addresses. SSSS is 16 bits for the organization to use for
subnetting. The remaining 64 bits are host address because for
some reason, the IETF thought that having really sparse host
assignments on a subnet was somehow valuable.
> You could probably do that in a year.
> Then what you do is get rid of the separate IPv4 and IPv6 numbering
> and replace them with a single numbering request that is used for
> both IPv4
> and IPv6. Then just adjust the justification requirements so that
> hardly has any, and IPv4 has harder requirements - that way,
> that only need IPv6 will just leave the IPv4 justification
> blank, and only get IPv6.
Given the significant differences in address structure, subnet size, and
implementation, i just don't see that as a feasible plan.
More information about the ARIN-PPML