[ppml] Policy Proposal: Decreasing Exponential Rationingof IPv4 IP Addresses
michael.dillon at bt.com
michael.dillon at bt.com
Sun Sep 2 12:00:58 EDT 2007
> A few = generally 2 or 3 years. Is that enough to call a trend?
If your data is monthly or weekly, then yes.
> > And the
> > changes have been explained due to macro events (CIDR introduction,
> > telecoms collapse) so even though the trends go through
> step changes,
> > they are still trends.
> If you can tell the trends from the macro events, please do
> so because as far as I know, nobody else has been able to.
I said changes were explained by macro events. In other words, pre-CIDR
there was a trend, then CIDR was introduced and the trend changed.
> > One could still do a worst-case scenario based on the pre telecoms
> > collapse trend
> I guess that would be going from 86 million in 2001 to 69
> million in 2002.
Pre telecoms collapse means *BEFORE* the telecoms collaps. Therefore I
was referring to the trend which existed *BEFORE* the drop in 2001-2002.
> > and then estimate the probability that macroeconomic events
> will lead
> > to that scenario. You then have an estimated runout date, and a
> > probability that it will occur.
> You can arrive at pretty much any outcome by just selecting
> the right start date.
But can you find macroeconomic explanations to reinforce your choice of
a starting date, and your choice of a trend? Unless you can tie your
numbers to macroeconomic triggers, they are just meaningless numbers.
> But 1995 was the first year with growth
> after the introduction of CIDR so we only have 12 years.
> It's just not enough to come up with something better than 5
> - 25 % growth per year = running out between Q3 2011 and Q3 2013.
Instead of playing amateur statistician, I suggest that you give the
data to a professional and see what they say.
> However, you can't predict a run on the bank, hoarding and
> all kinds of other interesting end games by looking at the past.
Of course not. You need to look at the macroeconomic factors. In the
case of a run on real banks, you need to look at the contracts that
banks have with their customers. They are not the same terms as
contracts in the 1920's and 30's. Then you need to look are reserve bank
systems and financial regulations in effect. All of these things allow
you to predict that there cannot be a run on the bank.
In the case of the RIRs, the fact that people must present technical
justification for IPv4 address requests pretty well prevents a run on
the RIR banks.
> Let me put it this way: would either of them care to publish
> an error margin to go along with their prediction? Or run the
> numbers on historical data and see how well the predictions
> fit what actually happened?
While this may be interesting, it does not change the fact that both
Tony and Geoff are using substantially the same data sources and
substantially the same methodology.
> I don't think there's another source of data.
IPv4 is used to address hosts. Hosts are computers. Computers are
purchased. Governments collect and publish statistics on purchases of
various types of things.
Companies consume IP addresses when they grow their network. It costs a
lot of money to grow a network. You need to have income to justify any
growth. Income is something that publicly traded companies must
And then there is RIR membership stats which might be used in
conjunction with corporate income reporting.
Router manufacturers tend to sell specialized routers for CPE. They
might be convinced to give a researcher access to statistics on CPE
Statistics and forecasting are disciplines which are taught at the
graduate level. Graduates are often given real-world problems to work
on. Professors of any given discipline tend to be chronically short of
ideas for such real-world statistical/forecasting problems. Etc.
More information about the ARIN-PPML