On 14 September 2012, the RIPE NCC essentially ran out of IPv4 addresses. They began to allocate IPv4 address space from the last /8 of IPv4 address with a very restrictive policy. Next in turn is ARIN. ARIN’s supply of IPv4 addresses is expected to last about 1 year.

Read RIPE’s announcement here

This week APNIC delegated their last IPv4 address the conventional way. The remaining pool of IPv4 addresses in the APNIC region will be delegated in small chunks of 1024 addresses. There is in other words nothing left other than breadcrumbs in this region.

In contradiction to my very exact estimate of the IANA depletion date, my tool and the mathematics I have been using failed to predict the APNIC depletion date with good accuracy. The algorithms that I have been using were not very good at predicting how an over 300% increase in demand over the last 2 months affected the depletion date. I would have been better off just using a linear algorithm with the last 2 months of demand as my input.

I must say that I was very surprised how quickly the APNIC pool got depleted. It appears that a “rush to the bank” happened once the members in the region realized that they might not get any additional IPv4 addresses. As the graph below indicates, the APNIC average burn rate went from slightly below 400,000 IPv4 addresses per day prior to the IANA exhaustion to almost 1.2 million IPv4 addresses per day after the IANA pool was depleted.

RIPE next

The current burn rates and the sizes of the RIR pools suggest that RIPE will be the next registrar to run out of free IPv4 addresses. RIPE have had a quite modest burn rate of about 150,000 addresses per day over the last year. The peak in the graph below in November is from when both Orange and T-mobile in the UK allocated 2 million addresses each (T-mobile and Orange actually merged their UK operations, so it is somewhat strange that they both could get 2 million addresses at the same day, but that is another story).

A linear extrapolation based on the RIPE burn rate and pool size suggests that RIPE would run out of IPv4 addresses in May of next year. There is currently no indication of a “rush to the bank” in the RIPE region. The current burn rate is actually slower than the yearly average.  

Modeling RIPE based on APNIC

We know that the demand in APNIC increased substantially when it was apparent that they would be the first RIR to be depleted. If RIPE is experiencing the same type of rush, when would then RIPE get depleted? Let’s do the math:

RIPE has 4.27 x /8 addresses left in their pool. This equals about 71,650,000 addresses.

The one year average burn rate in the RIPE region has been around 150,000 addresses per day. The rush experienced in APNIC increased the demand by 3.08 times. A similar increase in demand would get RIPE to a burn rate of 460,000 addresses per day. With such burn rate, RIPE would only be able to honor delegation requests for another 156 days.

Based on the calculations above, the RIPE depletion date could be as early as 19^th September 2011.

The wildcard in this calculation is of course if “RIR shopping” will become popular. The fear is that members of other regions will start allocating addresses in regions with remaining IPv4 addresses. If this will become a reality, we could see an even earlier RIPE depletion date.

Ripe is currently test announcing the following blocks: 185.0.0.0/10, 176.0.0.0/8, 37.0.0.0/10, 5.0.0.0/10. Those blocks show up as delegated in the statistics from RIPE, but they are not really used because they will be delegated the conventional way once the test completes. They falsely moved the RIPE depletion date earlier with 4 months.

These blocks have now been removed in the depletion calculations.

Now with the IANA pool getting depleted in a few moths it is interesting to take a look at what will happen afterwards. One source of IPv4 addresses that will come into use is the so called Various Pool. The history of the various pool is that prior to IPv4 classless Inter Domain Routing the class-B and class-C networks was allocated directly to end organizations from a set of IPv4 /8 blocks. These blocks are still around and got quite a bit of free IPv4 space in them (about 7.5 x /8). When looking at the IANA statistics file they show up as LEGACY blocks.

There is an agreement from 2008 between the NRO and IANA on how these blocks will be distributed amongst the RIRs. The idea is that they should be distributed to the RIRs according to this letter. The intention of the agreement appears to have been to distribute the blocks evenly. The XLS spreadsheet that is attached to the letter shows that they did a pretty good job making sure that all RIRs got the same number of addresses. 7.5 x /8 sounds good, right? However, there are issues with these blocks and things have happened since 2008. With the IPv4 depletion date around the corner, it is time to take a new look at the status of the various pool.

The 188/8 block
The 188/8 block was supposed to go to RIPE post IANA depletion. However, RIPE has already used up most of this block as I noted in March 2009

188/8 was completely empty except for one /16 allocating prior to RIPE starting to use it lately. Because RIPE already used up most of 188/8 they will get significantly lesser IPv4 addresses than the other RIRs when the various pool is divided between the RIRs.

The 191/8 block
The 191/8 block will go to LACNIC. An interesting thing with this block is that it is totally unused. The 191.255/16 network used to be reserved by RFC3330 but was freed up when RFC5735 was published in the beginning of 2010. So LACNIC gained 65k extra addresses, no big deal. But should this block still be in the various pool since it is empty?

The 196/8 block
It appears that the NRO and IANA forgot to include the 196/8 in their calculations. This block contains the equivalent to 0.71 x /8 or around 12 million IPv4 addresses in nice and large contiguous blocks. It has historically been used by several different RIRs but lately it delegations from this block has predominantly been made by AfriNIC who also handles the reverse DNS and whois for the block.

Personally, I’m all for giving less developed regions more IPv4 addresses. However, at some point it becomes very wasteful. AfriNIC already have (or will get) IPv4 addresses to cover their demand until the year of 2015 even with an exponential growth of the demand in the region. Five years from now I hope that IPv6 will be the predominant network protocol on the Internet and the value of an IPv4 address will be pretty low. Giving AfriNIC more IPv4 will not really help because no organization that hasn’t deployed IPv6 by 2015 will not be able to communicate with the IPv6 Only Internet that will be pretty big by then. It is unclear how NRO/IANA will clear up this mess. However, let’s assume for now that this block will be given to AfriNIC.

Gravel
The various pool are remains from /8 blocks used for class-B or class-C delegations prior to CIDR. They are therefore chopped up in smaller chunks and the free space is non-contiguous. Large service providers would rather get larger chunks of IPv4 addresses that small gravel to simplify maintenance and keep down the size of the routing table. Here is a list of the blocks that are in decent shape.
APNIC: 171/8 usable, 153/8 upper half free, 139/8 Some large free blocks
ARIN: 162/2 upper half mostly free, 172/8 mostly usable
LACNIC: 152/8 some usable blocks, 191/8 totally free
AfriNIC: 154/8 mostly free, 156/8 large chunks free on the upper half.
RIPE: 151/8 some large blocks free, with the biggest being 2 million addresses.

Conclusion
Unless something changes, the number of IPv4 addresses each RIR gets from the various pool will be:
RIPE: 0.67 x /8
LAC : 1.55 x /8
ARIN: 1.54 x /8
APNI: 1.54 x /8
AFRI: 2.25 x /8

The issues with 188/8, 191/8 and 196/8 as discussed above are the reason why the distribution is somewhat skewed. But how much “gravel” is there in those block? An interesting exercise is to remove all of the small blocks (smaller than 100k addresses). If we do that, the picture looks somewhat different:
RIPE: 0.33 x /8
LAC : 1.27 x /8
ARIN: 1.06 x /8
APNI: 1.27 x /8
AFRI: 2.01 x /8

As you can see, RIPE and ARIN appears to have gotten the most of the fragmented blocks from the various pool. If you are a large service provider you need to be aware of the fact that you might not be able to get large contiguous blocks of IPv4 addresses as we move closer to RIR depletion.

/written by Stephan Lagerholm (C) 2010

LACNIC recently allocated 1 Million addresses to NIC Mexico for further allocations to organisations in the region. This allocation pushes LACNIC’s pool a low level. We might see LACNIC requesting 2 more blocks from IANA soon. That allocation would most likely be the last allocation of IPv4 addresses LACNIC ever will make.

RIPE NCC and APNIC are also getting close to needing more addresses so we might see a total of 3 allocations of 2 blocks each soon. After the summer, I expect that we only will have 9 usable block in the IANA pool (except for the additional 5 that is already reserved for each RIR).

The estimated depletion date has been steady for the last couple of months. It is steadily pointing to a depletion of the IANA pool in March next year.

APNIC has handed out some large blocks lately to China and South Korea:
1. The network 183.192.0.0/10 of about 4 million addresses got allocated to a yet unknown entity in China. APNIC’s whois database is not yet updated to reflect this new allocation.
2. China Telecom Zhejiang province network allocated a /11 or about 2 million addresses.
3. Korea Telecom allocated a /11 or about 2 million addresses.
Those recent allocations along with the slow pace of allocations in Europe have changed how the end game most likely will look like. APNIC will most likely be the last RIR to request space from IANA. This last request will be for the last remaining odd block. I previously projected that RIPE would get the odd block. The remaining 21 x /8 that IANA got are now projected to be handed out like this: ARIN and RIPE gets 4 each, LACNIC gets 2 and APNIC gets 11.

France Telecom got 109.208.0.0 – 109.223.255.255 totaling around 1 million IPv4 addresses allocated today. This allocation will move the IANA depletion date earlier with about 10 days.

Quest recently allocated a /12 or about 500,000 addresses from ARIN. This is the first big allocation ARIN has made since the end of June when Embarq allocated a /12.

RIPE is still assigning surprisingly few addresses to their members pushing the depletion date forward slowly.

Ripe is getting closer to the anticipated allocation rate with a few more large allocations. Yesterday, Neuf Telecom in France allocated 2 million addresses (109.0.0.0 – 109.31.255.255). This bumped both the anticipated IANA run out date as well as the anticipated date RIPE will request more IPv4 addresses from IANA with 8 days.

Quite recently both Belgacom in Belgium and British Telecom in Great Britain got 1 million IPv4 addresses each from RIPE. This lowers the free pool of addresses that RIPE has and the free pool us quite low now. We can expect to see RIPE request an allocation of 2 x /8 from IANA soon.

The pace of the allocations in the RIPE region is currently defining when the IANA pool will get depleted (RIPE is expected to grab the last block from IANA). Due to lower than expected allocation rates in the RIPE region, this date have been pushed forward for quite some time now and the IANA depletion date is close to be pushed in to January of 2011.

Another interesting observation is that the allocation rate in the Middle East is much lower than I anticipated. Other than one allocation of 500,000 IPv4 addresses to Saudi Arabia in Febuary and 250,000 IPv4 addresses to United Arab Emirates in January, there is no substantial allocations made from that region during 2009. This is rather surprising, I was expecting more allocations from this region.