The real issue is that Geoff's linear projections against the current .75 /8's per month going out from the RIRs to hit a 2012 date don't match the historical growth.
I suppose I should respond here, particularly as the quote about using linear models is not a correct one. The projection model I use is updated daily at http://ipv4.potaroo.net using a rolling window of the past 1000 days to generate a predictive model of address consumption. Today's projection of IANA pool exhaustion is September 2011 and RIR pool exhaustion a little over a year later (assuming that the RIR pool can be cleaned out with 100% efficiency - which is an unrealistic assumption, of course). The growth model used is an exponential one, and the report shows the fit of linear, exponential and O(2) polynomial curves to the data used for projection (Figure 22). The choise of exponential is based on a decent least squares linear best fit to the logarithm of the smoothed data. I use a 1000 day baseline (i.e. the last 1000 days of hourly data) to produce the projection model. i.e. the model assumes that tomorrow will be a lot like today, and the changes will be the same changes that were evident in the past. The trend predictor I use in the growth in the advertised address range in the BGP table, and I derive RIR and IANA consumption figures from a combination of this primary trend plus a related trend in the growth of the unadvertised address pool relative to the growth in the advertised address pool I then model RIR behaviour in order to model IANA pool consumption in order to derive a date of pool exhaustion, using existing IANA to RIR address allocation procedures as the basis of the model. How good is this technique? I guess we'll know in about 6 years or so, but the nature of this daily update is that it will tend to self-correct over time. Late 2005 was a large 'bubble' of addresses entering the routing table - as happened early 2003. Currently the growth rate is lower than this recent peak. This makes predictor models a little more challenging in terms of figuring out whether there is any sense in artificially weighting more recent data over older data. The one thing I'll note here is that this model makes no consideration of any form of 'run' on remaining address resources, nor any consideration of a change in allocation practices, nor a change in industry demands over and above what's already visible in trend production. I notice that this thread is labelled "reality". Projections are not reality, they are always guesses to one extent or another as to what will happen. Reality is, of course, what has happened and what is happening. So, to some extent all of this predicting stuff is just fun with numbers. The reasonable high level bits to take away from this exercise and others that have occurred and will no doubt occur in the future, is that there is an increasing level of certainty that the current forms of access and distribution of IPv4 addresses will experience a discontinuity sometime in the next 4 - 8 years. regards, Geoff _______________________________________________ Ietf@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ietf