Re: [v6ops] Implementation Status of PREF64

[Owen DeLong <[e-mail]>]

> On Oct 13, 2021, at 10:34 , Gert Doering <[e-mail]> wrote:
> 
> Hi,
> 
> On Wed, Oct 13, 2021 at 10:23:28AM -0700, Owen DeLong wrote:
>>> Thinking we have arbitrary number of /64s to burn will have a hard landing
>>> when reality comes along.
>> 
>> Will it? Here???s reality???
>> 
>> More than enough /48s in 2000::/3 (1/8th of the total space) to hand out a few
>> hundred to every building now in existence or likely to ever exist.
> 
> You're overlooking intermediate hierarchies, and loss involved with that.

Nope… Not really…

Let’s divide ISPs into three categories — Small-ish, under 1M subscribers,
probably well served by a /28. Large (between 1M and 5M subscribers,
probably need a /24), and Huge — The Comcast/Verizon/etc. of the world,
who likely need a /12.

I’ll assert that there are probably 2-5 million, so let’s call it 5 million ISPs in the
small-ish category, a million or so in the large category, and maybe 200 world-
wide in the Huge category.

So that’s 5M /28s + 1M /24s + 200 /12s.

5M rounded up to a nibble boundary works out to 24 bits (16.7M), so 
subtracting from a /28, that gets us to a /4 which is 1/2 of 2000::/3.

Issuing a million (20 bits) /24s is another /4, so we’ve burned the other
half of 2000::/3.

Now we need to cover 200 /12s, which we can do from 4000::/3 and
still have 312 /12s to deal with any additional needs that may arise
and even if we burn through all of that, we’ve only used 25% of the
total address space.

> Can it be made to work?  Of course.

Easily.

> Can we have another quintillion for everybody's k8s clusters, and another
> one for every desktop PC with VMs on it?  Of course.

Quintillion addresses? Sure, 18 per. That’s a /64.

However, unless you expect an average building to exceed 50,000 such
combinations of devices, that still aggregates to a /48 for the building.

> Will we wake up to see that the start of this plan wasn't the best of ideas?

I doubt it. So far, the math seems to work.

> So, no, assigning whole /64s to large-numbered device types (unless they 
> serve as routers for one or more subnets, in which case I *do* support 
> delegating at least one standard-subnet-size per subnet) is not something
> I think we should do.  Balance needs to be found.

I think so far, there’s no evidence to suggest that /64 is not a good balance point.

The devices in question are, effectively routers, so it seems we agree.

Owen