[ppml] Combining Forecasts
tme at multicasttech.com
Wed Aug 29 07:38:44 EDT 2007
On Aug 29, 2007, at 6:21 AM, Iljitsch van Beijnum wrote:
> On 29-aug-2007, at 11:10, <michael.dillon at bt.com>
> <michael.dillon at bt.com> wrote:
>> Combining forecasts by averaging them has been shown to improve the
>> accuracy of forecasting. Perhaps it is time to dig some of the other
>> forecasts out of the woodwork so that we have more than just Tony's
>> Geoff's work.
I think that the trouble with the forecasts is not the numbers, nor
the math, nor even the approaches, but the assumptions. None of the
forecasts that I am aware of have any price elasticity in them, and
this will become critical as exhaustion nears. Let me explain this a
Of course an address market is developing and will develop, and that
- cause a land rush (which I see signs of developing now) and / or
- supplant the existing mechanisms.
Such developments are by their nature very hard to predict. However,
I can make some meta-predictions
- If nothing is done, there will be a land rush soon and once it
starts the remaining addresses will disappear very suddenly, probably
within a few months, starting at some time that is very hard if not
impossible to predict in advance, because this is a psychological
effect, much like stock market bubbles or crashes.
- If requirements are increasingly tightened, or artificial
restrictions are imposed like the exponential slowdown
of assignments proposed earlier, or the assignments are stopped for
"cooling off" once the land rush is underway, the market will
gradually (or rapidly) supplant the assignment mechanisms, and it is
conceivable that total exhaustion will not occur for some time. (At
some point, it would be easier to buy / lease / rent addresses than
to go to a RIR and obtain them through assignment, and so the
pressure on assignments would relax, and you would expect that
exhaustion would be approached asymptotically.)
- it is entirely possible that a land rush will cause a spurt of
market development and, as addresses start to become valuable, a lot
will "come out of the woodwork" as people start to realize that they
have some they can give up for cash, and then the system will convert
to an asymptotic approach to exhaustion.
Please note that I am not proposing a market here, I am just stating
that I see it as inevitable given current trends. Note that once it
forms (whether openly, or underground, or through some shadow play
like the long term lease of PA space from a consolidator) there will
be market forces opposing the adoption of IPv6 (as it will reduce the
market value of their assets). My personal opinion is that it would
be better to capture that market and de-nature it (say, by requiring
IPv6 support to trade PI space).
Is it useless to predict, then ? No, of course not. In fact, I think
that predictions should be used to help evaluate policy proposals in
this area (i.e., policy proposals should be included in predictions
to see if they actually have the desired effects). This, frankly,
sounds like a full time job for a professional statistician.
I would be glad to discuss thus further off list and kibitz about
ways to incorporate these ideas in forecasts.
> Here's mine. Please note that the numbers I'm about to mention are
> based on the dayly RIR allocation reports that I download from their
> FTP servers a few times a week. You can peruse this information
> yourself at http://www.bgpexpert.com/addrspace.php but caveats apply;
> also see http://www.bgpexpert.com/addrspace2005.php and http://
> www.bgpexpert.com/addrspace2006.php .
> On january first, 2005, the total free address space (= free in the
> global IANA pool and the space delegated to RIRs by IANA, but not to
> LIRs/ISPs/end-users by RIRs) was 97.4 /8s. A year later it was 87.5
> and on january first, 2007, it was 77.5. So we've been using up
> pretty much exactly 10 /8s a year in 2005 and 2006.
> Today the free space is 69.8 /8s. And interestingly, earlier this
> year a legacy /8 was returned to the IANA pool, something that hasn't
> happened for many years. So in 8 months (2/3s of a year) we've been
> using up either 7.7 or 8.7 /8s, depending on how you count that
> reclaimed block while 6.7 would have been expected during such a
> period based on the last two years. So that's a 15 to 30 % increase
> in yearly address use. Let's project this into the future, and
> include a stable 11 /8s a year option just for kicks:
> 0% 15% 30%
> used free used free used free
> 2008-1-1: 11.0 67.5 11.5 66.0 13.0 64.5
> 2009-1-1: 11.0 56.5 13.2 52.8 16.9 47.6
> 2010-1-1: 11.0 45.5 15.2 37.6 21.9 25.7
> 2011-1-1: 11.0 34.5 17.5 20.1 28.4 -2.7
> 2012-1-1: 11.0 23.5 20.1 0.0
> 2013-1-1: 11.0 12.5
> 2014-1-1: 11.0 1.5
> So if nothing happens, we have until about valentine's day 2014, but
> if we have a new trend on our hands it's either new year 2012 or late
> 2010. However, the yearly address consumption has never shown clear
> trends that keep going for more than a few years. What often happens
> is that there is a sudden increase from one year to the next, and
> then a decrease or stagnation in the year after. So I don't think
> we'll see a consistent 30% trend, or even a 15% one. I'd say there's
> a 25% chance we'll run out in 2011, a 40% chance that it's 2012 and a
> 25% chance it's 2013.
> You are receiving this message because you are subscribed to the
> ARIN Public Policy
> Mailing List (PPML at arin.net).
> Unsubscribe or manage your mailing list subscription at:
> http://lists.arin.net/mailman/listinfo/ppml Please contact the ARIN
> Member Services
> Help Desk at info at arin.net if you experience any issues.
More information about the ARIN-PPML