[arin-ppml] IPv6 Allocation Planning

[Owen DeLong]

In thinking about address planning and deployment, I've been doing some various
math.

It seems to me that we have more than enough addresses to use the following methodology
without any concern for exhaustion or shortage for a very long time. It also seems to me that
the advantages in terms of human factors and network planning simplicity are large enough
to be worthy of consideration.

As such, I'd like to get community input on whether they would support policy that would allow
ARIN to issue ISP allocations on this basis:

1.	Take your number of end-sites attached to each pop. Round up to
	the nearest value of x such that x = 2^n where n % 4 = 0 (in other
	words, round up your POP size to a nibble boundary).

2.	Take your expected number of POPs, say a Z-year plan (I seek community input
	on the desired value of Z, I'm thinking 5 years). Again, round up to
	a nibble boundary. Let's call this value y.

3.	Multiply x*y to get the number of /48s (ISP) needed. Convert this to a number (n)
	of bits (2^n=x*y).

4.	ARIN should approve you for a 48-n prefix or a /32, whichever is larger.

Examples:

3000 end sites per POP
154 POPs

3000 rounds up to 4096 (12 bits).
154 rounds up to 256 (8 bits).
Total need: 1,048,076 end sites (20 bits) 

48-20 = 28 -- This provider should receive a /28.

--

53,000 users per POP
350 POPs

53000 -> 65536 (16 bits)
350 -> 4096 (12 bits)
Total need: 268,435,456 /48s (28 bits)

48-28 = 20 -- This provider should receive a /20.

--

300 End sites per POP
10 POPs

300 -> 4096 (12 bits)
10 -> 16 (4 bits)

48 - 16 = 32 -- This provider should receive a /32.

I think the above examples are a reasonably representative set of values for medium, large, and small providers.

I realize that the resulting values result in a tremendous amount of extra space being issued as a result of the double rounding, but, I believe it is worth while for the likely gains in aggregation.

To put it in perspective, if we took twice as many ISPs in each region as there are active ASNs in the entire IPv4 internet today, and gave them all /28s, we would barely use up the first /12 in each region. We would still have tremendous room to grow within the first 11 /12s out of 512 that are contained within 2000::/3.

In my estimation, the average allocation under this scheme would likely work out very close to that, but, even if I'm off by a factor of 16, we'd still only use most of 176 /12s and we'd have 336 /12s untouched in 2000::/3 with the internet being more than double its current size.

Please let me know what you think of this idea. If it receives some positive feedback, I'll start working on turning it into policy.

Thanks,

Owen