[ppml] And as for assignments...

On 25-aug-2007, at 21:58, Jonathan Barker wrote:

> But, once upon a time, IPv4 didn't have classless routing  
> capabilities.
> That was built in later. If there's demand, all these other  
> features can
> surely be built in for smaller blocks as well

Yes, and that's our insurance policy against running out of IPv6  
addresses: when we make a mess of the upper 64 bits, we can slice and  
dice the lower 64 bits and still have a few billon times the address  
space of the IPv4 internet.

> I suppose I've held a little too much conservatism over from IPv4
> constraints, as I have been very constrained when it comes to IP
> addressing. We're the most IPv4 efficient ISP I know of - but I just
> cannot understand why a single family dwelling would ever need or want
> 2^64 addresses. It's just so phenomenally wasteful.

You have to remember that IPv6 was designed a decade and a half ago,  
when things were very different from the way they are now. For  
instance, unless I'm mistaken, DHCP wasn't standardized until late  
1993 and didn't come into wide use until some years later. So when  
the decisions about how IPv6 works were made, many, if not most, IPv4  
systems were still configured with an address manually. We didn't  
have ethernet switching, or even the notion that at some point pretty  
much everything would run over some flavor of ethernet. So subnetting  
was necessary in many more cases than now. Also, other protocols than  
IP were still in common use. So it's not strange that IPv6 was  
outfitted with something very similar to the inclusion of the MAC  
address in the network layer address like IPX and CLNP do. (Although  
the IPv6 designers got a bit carried away and pretty much included  
ALL of the address configuration mechanisms present in ALL protocols  
in IPv6.) The advantage of including the MAC address is that a system  
gets the same address every time without the need to do complex stuff  
in a server to make this possible.

If we were to design IPv6 today, it's certainly possible that we'd  
use 64-bit addresses with a /48 per end-user which can then  
accommodate 16 4096-address subnets or 256 256-address subnets or  
some other combination.

Although it would be better to use variable length addresses. This  
has the advantage that we can be economic with address bits at first  
and deploy more of them as required. This could give us for IPvX what  
NAT gives us today: the ability to plug in a box, get an address, and  
then turn that address into multiple addresses for further downstream  
delegation. But then in an architecturally clean way.

-- 
I've written another book! http://www.runningipv6.net/