I'm going to present the following policy change at the ARIN
meeting in April. In a few days I have to submit the exact wording
to ARIN and I'm looking for comments before I do this.
Basically this policy loosens the rigid 80% utilization
requirement in a progressive fashion to recognize the
inevitable overhead of hierarchy in larger networks.
1. All requests for additional IPv4 address space shall require the
efficient utilization of the sum total of all existing allocations
including all space reassigned to customers, if any.
2. The HD(Host Density) ratio of the sum total of all previous allocations
shall be greater than or equal to .966 and the HD ratio of the most
recent allocation shall be greater than or equal to .930 in order to
receive additional space.
3. The HD ratio is calculated as log(utilized IPv4 addresses) divided
by log(total addresses in all previous allocations). In this formula,
log refers to the natural logarithm.
The HD ratio was proposed as a way to determine allocation usage
for IPv6 address allocations. For more details on this, please refer to
RFC 3194 <http://www.faqs.org/rfcs/rfc3194.html>. There is some detailed
background discussion about applying the HD ratio to IPv4 allocations in a
proposal by Paul Wilson posted to the APNIC mailing list on Aug 7, 2003
and he presented the it to the annual APNIC policy meeting using these
I am not suggesting that ARIN should adopt the APNIC proposal and although
Paul invents a new name for the HD ratio, I prefer to keep the original
The basic thrust of this proposal is to replace the rigid 80% usage
by the more flexible HD ratio and to shift the emphasis away from the last
allocated block to include the total allocated address space. To that end,
the .930 criterion for the last block is a lot looser than the existing
requirements for the last block. This is because the utilization threshold
establishes a time buffer between the beginning of an ARIN application for
additional addresses and the final deployment of new addresses in the
operational network. By using a looser criterion as network size grows, we
are also expanding this time buffer. This recognizes that the economy is
more dependent than ever on the smooth running of our networks and we
should not artificially force larger members to operate with virtually
no safety buffers for implementing new addresses. This safety buffer size
is important because larger networks have more involved processes for
changes to their network and these processes take time.
Paul Wilson's paper contains ample discussions of the technical
for using the HD ratio. I have proposed that we use the .966 number that
suggests, I believe there may be valid arguments for reducing this
perhaps to .960.