[6bone] Is minimum allocation /64 now?

[Frederick Bruckman]

On Sun, 26 Oct 2003, Dan Reeder wrote:

> The problem is that, perhaps because some of us have had to live under the
> strong arm of apnic, that the tendency to want to conserve addressing is a
> bit of a habit. Personally whenever I see things like /48s being given to
> users left right and center I get reminded of the consequences of Stanford
> being given a v4 /8 way back in the early days. It just reeks of
> wastefulness. Just because we can, and just because some (antiquated?)
> documents say so, does that mean we should?

The problem was, that there turned out not to be enough addresses
for the Internet as it came to be, period. CIDR, and use of formerly
reserved address spaces, are consequences of that simple fact. Making
the orignal allocations denser would not have prevented the problem.

> To me, the whole problem is a bit of a "chicken and egg" cycle. The rfcs and
> powers that be say /64 is the minimum primarily because other rfcs have
> dictated addressing schemes, and that the autoconfiguration software doesn't
> support network prefixes in greater length than /64. But then the
> autoconfiguration software developers say they only support  up to /64
> beacuse of rfcs!
> Why can't someone bite the bullet and just develop a daemon like radvd that
> will simply use pretty much any prefix length thrown at it? I've got a /64
> on my lan here. If the advertisement software supported it operationally
> speaking it would make absolutely ZERO difference if I changed it to /80...
> or /112 or even a /120. And I bet it would make almost zero difference to
> the majority of the readers on this list (i'm not really talking about ISP
> network operations/addressing here though)

Uh, no. You don't get it. The lower 64-bits are for your globally
unique host address. Allowing as many bits for the host addresses as
for the network addresses obviously means that IPv4 CIDR will never be
repeated for IPv6, and that therefore router manufacturers are free to
bake that assumption into the hardware.

I think what the Powers That Be need to do, to stop this topic from
coming up endlessly, is to change the marketing language from "IPv6
gives you <inconceivable number that there is no word for> network
addresses," to, "IPv6 gives you 2^64 == 16 quintillion networks
(American usage), and an unlimited number of hosts on each network".

It would be smart policy to give each physical location a 48. Most of
the 65,536 networks will be "wasted", if you will, but consider where
the room for expansion is likely to be needed. Will some users want
more than 16 or 256 networks? (Yes.) Or will there be more than 64
trillion locations? (Not likely.) Points of presence that need *more*
that 65,536 networks can simply use the same equipment and methodology
that an ISP uses, so no problem there either.

Frederick