[ppml] Combining Forecasts

On 30-aug-2007, at 15:51, John Curran wrote:

> I doubt if a few months of uncertainty either way would
> matter for purposes of policy consideration, whereas a
> difference of a few years is definitely relevant since one
> wants to calibrate any policy changes against the need.

No, one doesn't. The predictions are too inaccurate to base policy  
on. Just a few years ago, Geoff Huston told us IPv4 would last well  
into the 2020s. Now he says 2011. Both are reasonable projections  
based on the data available at the time, but I have little doubt that  
in another couple of years, things will look different once again.  
However, by that time we'll be that much closer to running out so  
maybe that one will stick.

> For example, if we're looking at 2015 as an expected IPv4
> depletion date, it's possible that IPv6 could have a sizable
> deployed base by then and be fairly mature

Or people saw that things weren't that dire with IPv4 and shelved  
their IPv6 plans.

> If one considers a different scenario with a 2012 depletion
> date, we've got a different situation and need to explore how
> to increase conservation and encourage IPv6 deployment.

Increasing conservation is bad for people who need new IPv4 space. If  
it's bad enough, this may get them to move to IPv6, so then the net  
effect would be positive. But it's also possible that they stick with  
IPv4 and the bad effects translate in higher prices and fewer options  
for end-users, while the transition to IPv6 needs to happen a few  
years later anyway.

> Using an IPv4 depletion date of 2009 strongly suggests looking
> into aggressive conservation,

Depletion by 2009 means 600 million addreses per year are used up. I  
think when we reach that, it's too late for conservation.

> proactive IPv6 deployment,

May want to do that regardless of the predictions.

> and reclamation/reuse of legacy space,

You need time for that, and at 600 million/year it buys you next to  
nothing. It's in the opposite situation, where we only burn 150  
million addresses/year, where reclamation could be useful.