[ppml] Policy Proposal 2007-16: Ipv4 Soft Landing - a simulationanalysis

[Geoff Huston]

Bill Darte wrote:
> Geoff,
> 
> Thank you very much for the input and analysis.
> 
> It seems that the IPv4 Soft Landing proposal makes little difference given your assumptions and calculations.
> 
> Could you comment on what assumptions might be changed that WOULD materially alter the dates of exhaustion and why such assumptions weren't made?
> 
> What I ask seems simple to me, but if the request turns out NOT to be simple or would be a tremendous amount of work, then I withdraw.
> 
> Thanks again for all the work you have done for the policy process at ARIN and elsewhere!
> 
> Bill Darte
> ARIN AC

The key assumptions I've made are:

1. For requests for additional IPv4 address space the requestor has 
fully utilized their existing holdings at the point of the additional 
allocation of address space.

2. The distribution of new requestors who are requesting initial 
allocations and requestors wishing a further allocation are largely 
unchanged from that of the average of the past three years

3. That the demand model is uniformly spread across the year.

4. That the distribution of size of address requests is unchanged from 
that of the average of the past three years.

5. That the unadvertised pool of addresses has the same factors acting 
on it that have been visible in the past 3 years - i.e. no change in 
policies here.

About the only assumption I know that is demonstrably false is 3. and 
the demand model is not at all uniform, and in looking over the data I 
suspect that the 3 year baseline I'm using is too short, and a 5 or 6 
year baseline for the projection model might provide a better trend for 
projection.

If you look at Figure 27f of http://ipv4.potaroo.net the post 2000 total 
demand model of RIR allocations shows a 7 year trend of growing by 
around 1.2 /8s per year, from 3 /8's per year in 2000 to 12 /8s per year 
by the end of 2007. Even using a linear projection this gives us around 
3 years before the remaining 42 /8s are exhausted.

Now this is predicated on the assumption that we over-service the demand 
model, and for each deployed address there is a oversupply factor of a 
further .25 of an address (the 80% rule). Now David's proposal wants to 
progressively drop this oversupply factor to 0.18 (85%) and then to 0.11 
(90%). But this is not a uniform distribution. The matching of a end 
user population to an allocated prefix essentially quantises the 
population using a logarithmic division. i.e. if the range of end user 
populations for each allocation are distribution between 100 (/24) and 
1.3M (/8) then a /8 covers the largest half of the populations, a /9 the 
next quarter, a /10 the next eighth, and so on. So if you change the 
oversupply factor in this linear fashion this does not directly 
correlate to a reduction in total address demand by a comparable amount 
because of the logarithmic nature of the quantisation of demands into 
powers of 2 to form an address prefix. So I'm not sure that these 
efforts to change the 80% utilization factor in IPv4 have any 
significant impact  ("significant" in terms of altering the address 
consumption rate by 20% or more).

On the other hand the short term indicators are also important, and over 
the past 6 months the total RIR allocation rate has dropped from an 
equivalent rate of some 14 /8s per year to the current allocation rate 
of 7 /8s per year. This is strongly evident in APNIC, RIPE NCC and 
LACNIC where the most recent allocation rate is half that of the rate 
earlier in 2007. While its pretty clear from the graphs that these 
variations in the rate have been present in the past, the periodicity is 
not clearly aligned to an annual cycle, so the drivers that cause surges 
in demand for addresses from regional groups are not solely based on 
annual business cycles. Beyond that observation I'm not sure where this 
line of thought leads in terms of being able to add this factor into the 
  address consumption model that I use, nor am I confident that even 
adding this additional volatility in demand into the model would 
materially alter the outcomes given that below the volatility is a 
multi-year long steady growth in demand that is evident in all 5 RIRs.

So, in conclusion, I'm not sure that I could offer you a variation of 
assumptions in this model that would have a major impact on the 
cumulative address consumption projections.

regards,

   Geoff