[ppml] Use of HD ratios (esp. 0.94)

[briand at ca.afilias.info]

Having read more about HD ratios (RFC 3194), and re-reading the current NPRM,
and proposed changes, I think there needs to be further discussion on the
matter.

Let me first present some math stuff, since it was confusing to me, and I
have three math degrees from the University of Waterloo (;-)):

HD = log(things)/log(max-things)

Which, after much complicated stuff, gets rearranged into:

things = (max-things)^HD

But, we are used to max-things of 2^N, so we get:

things = (2^N)^HD = 2^(N*HD)

Here's where I think we stared getting messed up:

The current and proposed texts, all are using, not true "HD", but
rather, HD-56, where the base unit of measure is /56, or 2^56.

In terms of allocation efficiency, and utilization requirements, using
HD-56 instead of some other kind of HD, unfairly penalizes those who
act as good "Internet citizens", and allocate longer prefixes to customers.

Consider two comparative situations. ISP A, gives out /48's, and ISP B,
who gives out /56's. Each does so from an initial allocation of /32.

ISP A can give out at most 65536 prefixes. ISP B can give out ~16M.
The respective thresholds would be ~24K vs ~6M.

However, at review time, both have a threshold of 36.9% utilization.
The presumption is that ISP A's assignments are 100% utilized.

If instead, the HD ratio was based on a value related to the prefix
sizes used in allocations, then ISP A would have to have a utilization
level of 51.4%.

I don't know for sure how we should handle VLSM assignment space rather
than fixed-size assignments.

But I do believe we should be rewarding efficient use, by basing the
HD ratio on the base assignment size.

What do folks think?

Or am I mis-interpreting how HD-56 is to be used? Are we supposed to count
each object, regardless of size, as 1, and the utilization percentage
would then be based on object count vs threshold values?

If the latter is the case, then a /32 would only be able to give out
at most ~47K /48's, and the rest would have to be /56 or longer,
for a total of ~6M, before justifying another /32.

Can I ask anyone from ARIN to comment on the *intent* of HD and HD-56,
for evaluating IPv6 assignments?

Brian Dickson