[ppml] IPv6>>32

[Paul Vixie]

jordi wrote, in response to leo:

> Apart for technical issues, I will say that is too late, years late,
> to change the host length.
> 
> There has been a strong effort in coding and deployment. May be not so
> much in US, but definitively in AP and Europe, which can't be changed
> so easily.

with respect, i don't see this as directly relevant.  i am aware of sites
that use static addressing or dhcp6 because of various perceived problems
with EUI64 autoconfiguration.  i am therefore certain that EUI64 can be
gone without in at least some situations.  therefore an RIR policy of the
form "if you can live without EUI64, you can have PI space" would likely
have significant uptake.

> On the other side, you're probably missing the picture of the hardware
> issues, I mean silicon implementations of IPv6, which mean a huge
> investment and years to be replaced in the market.

i'm still waiting for someone to identify a router that either works better
or goes faster or costs less or can handle more subnets if you use /64's
per LAN rather than /96's.  my own testing with M20's and 7206VXR's says
that it makes no difference, but i know there are more modern routers and
i'm waiting for someone to identify one that cares about your prefix size.
("one" might not be enough to sway my opinion, but it'd be a beginning.)

> I don't really think that's possible, and probably all those are the
> reasons why the people that you could have asked in IETF have in mind
> to avoid going into this direction.

the people actively involved in designing protocols via the IETF just don't
care about economics, as a rule.  "what it would look like it deployed" was
not a consideration with classful IPv4 ("how are we going to *route* 16
million class c's?"), or DNSSEC ("if a zone can't be secured until its
parent is secured, and if the parent has no incentive, *then* what?") or
in IPv6 ("we've got 128 bits, let's burn half of them on something nobody
cares about that has unforeseen privacy effects"), or A6/DNAME ("let's put
*even* *more* pressure on the size of the global routing table, shall we?").

> Instead, I feel much more realistic to reconsider the HD-ratio for
> IPv6.

i think that this should (and will) also be done.

> Final consideration, do you really believe in 50-100 years we will
> still be using IP at all ?.

my final consideration is, since the size of the ipv6 installed base is very
much smaller and very much more fluid/serviceable than the size of the ipv4
installed base, why are we assuming we can forklift-upgrade the latter but
not the former?

paul