[ppml] IPv6 addressing, allocation, and subnets

Brian Dickson briand at ca.afilias.info
Fri Nov 16 16:53:45 EST 2007

Tony Hain wrote:
> Brian,
> Clearly you have yourself wedged into the mindset where you don't want to
> deploy /64 subnets. That is fine, and no one is requiring you to. It is also
> clear that smart people can come up with cases where the existing IPv6
> design is not optimal. The error of their ways though is when they try to
> turn these corner cases into something bigger than they are. 
I don't believe you remember the discussions we had at NANOG where I 
indicated that none of what is being discussed, is related to plans 
involving me or my employer.
(We do in fact deploy only /64 subnets, except for external 
point-to-point links to ISPs.)

I am addressing the issues that affect those who have not yet deployed 
IPv6, and for whom the deployment is likely to be fraught with 
difficulty and peril, specifically *because* of those who view IPv6 as a 
universe where all prefixes are /64.

It is important to adopt policies which support the needs of all ARIN 
members, *and* their customers, not just the majority of ARIN 

The fact that someone has not yet deployed IPv6, does not mean that that 
party's needs are irrelevant.
We should plan to accommodate the needs of those who show up late to the 
ball, rather than punish them for not being part of the "in" crowd.

The whole point of the Internet is connectivity, and facilitating other 
parties connection to the Internet is part and parcel of, and the entire 
value of, being on the Internet.
> http://www.isoc-au.org.au/06ipv6summit/talks/Ron_Broersma.pdf  shows on
> slide 22 that trivial mappings between IPv4 & IPv6 are possible, and the
Uh, no it does not.

I would appreciate it if you would at least read what I wrote, or if 
presenting counter-arguments,
read things to which you point people.

For the benefit of those too busy to follow your URL, here's the text 
from the page:

    Example Re-addressing scheme
    Re-address the network for consistency between protocols
    – IPv4 – move all subnets to /24 or larger
    – Align VLAN number with 3rd octet of IPv4 address
    – Align IPv6 “subnet number” with the above
    [pictures omitted for brevity]
    – Reduction in complexity
    – Easier for operations staff, once re-addressing is complete
    – Assumes you have enough IPv4 address space to change it as well.

So, this is *not* a mapping scheme - it's a combined addressing scheme, 
which requires that all IPv4 subnets be *renumbered* into /24, and that 
only /24:/64 mappings are supported.
And, it requires that you have *space* to do this renumbering in IPv4.

None of this meets the implied situation that mappings *are* trivial.

Rather, it shows that *only* trivial mappings scale.
> You are arguing to redesign the protocol in the wrong forum, and the wrong
> decade. The time to have made your points was 1995, and even then similar
> proposals were considered and rejected. It is time to let this bad idea
> go...
CIDR and VLSM are hardly bad ideas, and support for analogous schemes 
between IPv4 and IPv6 is very timely.

The fact that widescale deployment of CIDR happened concurrently or 
after IPv6 design time (1995) is the counter-argument. It is not 
surprising that, absent experience with IPv4 CIDR, that IPv6's authors 
didn't see its value.

IPv6 is not what the authors of the RFCs say it is outside of the RFCs 
themselves. It is only what the RFCs themselves dictate - and from the 
main IPv6 RFC, it has *no* internal structure.
Some uses of IPv6, and some protocols, assign meaning to certain bits.
And use of IPv6 which is consistent with the RFCs, is neither redesign, 
nor incorrect use of, the protocol.

It is time to get with the CIDR program, not to advocate the antiquated 
class-ful paradigm.


Brian Dickson

More information about the ARIN-PPML mailing list