Orbit

At 12:13 PM 17/03/2006, Save Vocea wrote:

Dear colleagues

APNIC welcomes comments and suggestions on the following policy proposal:

---

Status of [prop-020-v001] "Application of the HD ratio to IPv4"

I would like to request that the following analysis be considered in relation to this policy proposal

Thanks,

Geoff Huston APNIC

An Analysis of the Sensitivity of using the HD Ratio for IPv4 Address Allocations

Geoff Huston V1.0 22 February 2005

This document describes the outcomes of an analytical process intended to describe the sensitivity of the use of HD Ratio metrics as the means of assessing address utilization efficiency, and the relation between the use of HD Ratio values and projected lifetimes of the unallocated IPv4 address pool.

This study uses RIPE allocation data as the base of the analysis. As noted in this document a similar analysis relating to APNIC IPv4 address allocations indicates that the total address consumption by APNIC is this proposal were to be adopted would be an increase of 47%. As noted this has some significant implications for IPv4 lifetime projections for the remaining unallocated address pools. I also note that the major change here is the address allocation sizes for larger allocations under this policy, while allocations for smaller blocks would be substantially similar to the current 80% utilization policy.

1. Methodology --------------

The methodology used here uses only published RIR allocation data. The primary data source for RIPE NCC data is the delegated file: ftp://ftp.ripe.net/pub/stats/ripencc/delegated-ripencc-latest

All IPv4 allocation records with an allocation date on or after 1-Jan-2000 are collected. The allocation sizes are rounded up to the next largest power of 2, or 256, which is the greatest. The relative proportion of each allocation size is also calculated. This is shown in the table below (Table 1).

---------------------------------------------------------------- Table 1 - RIPE NCC IPV4 Address Allocations (since 1-Jan-2000)

Size Number Relative Cumulative Frequency Relative Frequency

/24 2637 23.04 23.04 /23 1383 12.09 35.13 /22 934 8.16 43.29 /21 545 4.76 48.06 /20 2247 19.64 67.69 /19 1713 14.97 82.66 /18 784 6.85 89.51 /17 407 3.56 93.07 /16 499 4.36 97.43 /15 135 1.18 98.61 /14 75 0.66 99.27 /13 44 0.38 99.65 /12 21 0.18 99.83 /11 15 0.13 99.97 /10 4 0.03 100.00 ----------------------------------------------------------------

The assumption made here is that these allocations are made under a policy of a uniform 80% utilization efficiency. From this can be calculated the inferred maximum end use count for each prefix size (Table 2).

---------------------------------------------------------------- Table 2 - Inferred Maximum End Population Count for each Prefix Size under the uniform 80% efficiency policy

/24 205 /23 410 /22 819 /21 1638 /20 3277 /19 6554 /18 13107 /17 26214 /16 52429 /15 104858 /14 209715 /13 419430 /12 838861 /11 1677722 /10 3355443 /9 6710886 /8 13421773 ----------------------------------------------------------------

The HD ratio is calculated by the function: HD = log(used)/log(addresses). This implies that the population can be inferred for any given prefix size using the equation: used = 10**(HD x log_base_10(addresses). The inferred maximum end use count for each prefix size using an HD Ratio value of 0.96 is shown below (Table 3).

---------------------------------------------------------------- Table 3 - Inferred Maximum End Population Count for each Prefix Size under an HD = 0.96 allocation policy

/24 205 /23 399 /22 776 /21 1510 /20 2937 /19 5713 /18 11113 /17 21619 /16 42055 /15 81811 /14 159147 /13 309590 /12 602249 /11 1171560 /10 2279048 /9 4433455 /8 8624444 ----------------------------------------------------------------

The next step is to determine the relative impact on address consumption by changing from a uniform 80% utilization efficiency metric to one determined by an HD Ratio setting of 0.96.

To do this a sequence of 10,000 allocations are simulated. with each allocation being in the range of a /24 to a /10 prefix. with a probability of any particular prefix being selected based on the relatively frequency distribution of Table 1. The inferred population lies between the maximum population of this prefix and that of the population of the next smaller prefix in Table 2. A random value is drawn from this population range (this is a uniform probability selection between the two extreme population values, so that any population value is equally likely to be selected). This population value is used as a lookup key into Table 3, and the next highest population count is used to determine the equivalent HD Ratio allocated prefix. In effect, this approach generates a series of demand populations that would generate the existing RIR allocation prefix distribution, and then uses this population set to generate a HD-Ratio- based set of allocations that would correspond to this population distribution. The total amount of allocated address space is calculated in each case, and the ratio of the two address pool sizes is recorded.

This experiment has been repeated 1,000 times in order to determine a stable average value for the relative increase in address consumption corresponding to a change in the address allocation policies from uniform 80% to an HD Ratio of 0.96, assuming constant demand for addresses.

This relative change in address demands can then be added into the IPv4 address consumption projection (see http://ipv4.potaroo.net). The change here is in the simulation of the address consumption model, where in the base model all RIR's are assumed to be operating a uniform address efficiency metric of a uniform 80% utilization target. The same exponential growth model in advertised address growth is used, but this model is augmented by the relative increase in address consumption as contributed by the HD Ratio allocation metric. The unadvertised address ratio is then derived from this higher advertised address count, and this, in turn, generates a more rapid overall address consumption model. The measure under investigation in this case is the change in predicted date of the exhaustion of the IANA unallocated address pool

2. Results ---------

The relative distribution of allocated prefixes by the RIPE NCC using an HD Ratio of 0.96 as an allocation efficiency metric would be as shown in Table 4.

---------------------------------------------------------------- Table 4 - RIPE NCC IPV4 Address Allocations

Size 2000-2006 HD Ratio Relative Relative Frequency Frequency

/24 23.04 23.23 /23 12.09 11.37 /22 8.16 7.87 /21 4.76 4.85 /20 19.64 16.33 /19 14.97 15.21 /18 6.85 8.58 /17 3.56 4.39 /16 4.36 3.88 /15 1.18 2.39 /14 0.66 0.86 /13 0.38 0.50 /12 0.18 0.28 /11 0.13 0.15 /10 0.03 0.09 /9 0.00 0.02 /8 0.00 0.00 ----------------------------------------------------------------

From the simulations of registry allocations, the use of an HD Ratio of 0.96 for IPv4 address allocations made by the RIPE NCC is predicted to increase total address consumption by 46% over the existing flat 80% utilization allocation policy framework.

The current prediction for the data of exhaustion of the IANA unallocated address pool is 12 January 2012, assuming, among other factors, a continued application of the constant 80% address utilization metric. If the RIPE NCC were to adopt an allocation policy of using an HD Ratio of 0.96 to access IPv4 address allocations, and no other changes were made to the mode, and no other RIRs were to adopt such a policy to use the HD Ratio as a utilization metric, then the impact on the predicted exhaustion date is an overall change in address consumption rates by approximately 17% (as the RIPE NCC is responsible for some 38% of all allocated IPv4 addresses), and a predicted unallocated IANA pool exhaustion date of 9 December 2010 under these conditions (or approximately 1 year earlier than the predictions using the current address allocation policy framework

A related consideration is that of the adoption of such a policy proposal by all 5 RIRs. If this were the case, and the adoption of this policy was to be effective immediately, then the relative increase in overall address consumption for each RIR would be: Afrinic 39%, APNIC 47%, ARIN 46%, LACNIC 47%. The simulation of IPv4 address consumption under these conditions predicts that the IANA pool of unallocated addresses would be exhausted by 22 March 2010 (or approximately 2 years earlier than the predictions using the current address allocation policy framework).