[ppml] Proposed Policy: Proposal to amend ARIN IPv6 assignment and utilisation requirement

[Rich Emmings]

On Thu, 8 Sep 2005, David Conrad wrote:

>
> In theory, true.  In reality, I suspect developers will take shortcuts where 
> they can.  If there is a commandment that "thou shalt always have /48 or 
> shorter", I suspect retrofitting longer prefixes will break lots of embedded 
> stuff.

One of the things that I feel will be true.  CIDR or not in IPv4, there's a 
lot of classful legacy code out there that is still alive and living in your 
network, even in classless protocols.  I can see 32 and/or 48 bit memory 
locations showing up in hardware based FIB's with some sort of punt when 
it's not an expected boundary.

>
>> There's a lot of space out there.  I don't think many people fathom the
>> concepts of 2^128.
>
> And I don't think people understand that IPv6 is not really 2^128. 
> Currently, it is between 2^48 and 2^64.  Yes, both are still very, very large 
> numbers however the RIRs are, even at this very early stage of deployment and 
> before the promises of ubiquitous Internet connected devices is anywhere near 
> reality, allocating /19s and /20s according to current policies.  Obviously, 
> there are the same number of /19s and /20s in IPv6 as there are in IPv4.
>

Agreed on both, and one of the things I try and remember is to subtract 
that /64 from the /prefix to get real numbers.  Nevertheless, 2^64 is not 
only twice 2^32, and even with the reserved space is removed, there's still 
a very large amount of space out there.  Nevertheless, pending rewrite of 
autoconfiguration, a /64 is what's gone.

Even with that, 2^64 is a very large number -- if it was milliseconds, I'm 
not sure the universe has been around that long.

To some extent, the proposal is a proponent of more free space available for 
the future, vs more free space available now.  Someone with a /56 may 
possibly grow to a larger block or set of /56's.  I would intuit that a /48 
likely means single assignments for most everyone.  Maybe the 0.001% 
mentioned elsewhere is the difference between /56's and /48's.

>
>> No system will be perfect, we have room to adjust it later, and there's
>> enough space out there that we're not at risk of running out over the short
>> term.
>> 
>> Perhaps table things for a year, or, until we need to think about issuing
>> 4000::/3, and see what real world experiences suggest then?
>
> As a person who has personally and repeatedly experienced first hand the 
> repercussions of "historical inequities" in IPv4 addressing, I'm not sure it 
> makes sense to repeat that particular IPv4 mistake.

I've seen that too, however, we're less than 50% on 2000::/3 and haven't 
broached into the other 6 /3's that would be available at some point.  We 
have more room to make adjustments than with IPv4 space.  If the finite 
space was just 2000::/3 and were approaching 50%, allocation size would be a 
more important factor than getting things more set in concrete.

I run into issues with rollout related to "Well, they're still changing 
things, let's wait"   I'm willing to accept a less than perfect initial 
system that has room to be moved into a more perfect state at a later date.