The distribution of IP addresses has always occurred in a somewhat socialist manner: "to each according to his needs." But those days are drawing to an end now. So far in 2012 within the North American region, no less than a quarter of all IPv4 address blocks obtained by new owners/users have been traded in some way. This is despite the fact that ARIN (American Registry for Internet Numbers)—the Regional Internet Registry (RIR) that supplies North America with IP addresses—still has at least 40 million IPv4 addresses available by one count. More than 100 million by another.
Researchers at Syracuse University in New York and the Delft University of Technology in the Netherlands went over the records of the three RIRs (of five) that allow "transfers" from an organization holding address space to another organization in need of address space. Standard operating procedure is for the former to return addresses to the RIR governing their region, and for the latter to obtain addresses from that RIR. However, in the Asia-Pacific region this is no longer possible: APNIC has less than 17 million IPv4 addresses left, and each organization can now only get one last block of 1,024 addresses. Last week, the RIPE NCC (Réseaux IP Européens Network Coordination Centre) in Europe found itself in the same situation—but the study was completed before the RIPE NCC ran out of addresses.
So in Asia, the Pacific, and Australia, the only way to get a significant amount of address space is to find someone with addresses to spare and coax them to transfer some to you. Presumably, this entails some level of monetary compensation. So researchers looked at the period from 2009 until the middle of 2012. In the RIPE and ARIN regions, IPv4 addresses were still available to those who can show they really need them. In light of that, you'd expect to see transfers in the APNIC region, but not in the RIPE and ARIN ones.
And indeed, in the APNIC region, 191,744 addresses were transferred in 2011 with another 713,216 in the first half of 2012. In the RIPE region (Europe, the former Soviet Union, and the Middle East), the researchers couldn't find any address transfers. But surprisingly, in the ARIN region (North America)—where there is no immediate shortage—no less than 821,504 addresses were transferred in 2011 with 4.22 million in the first half of 2012.
In the paper titled Dimensioning the Elephant: An Empirical Analysis of the IPv4 Number Market, the researchers point out this goes directly against a prediction made by yours truly last year: "at the ISP level, a functioning market won't form at all, or will break down very quickly after it forms." The lack of address transfers in Europe backs this up. So does the fact that in the APNIC region, despite being in "you only get a final block of 1,024 addresses" mode, only 0.7 million addresses were transferred.
But in North America, a transfer market did form. In most cases, nothing is known about the prices paid for address blocks. In other instances this information has come out, with the price settling around $10 per address. Why would organizations buy addresses rather than get them for (almost) nothing from ARIN? (I mean, the researchers even note that in one case, the transfer price would pay for the yearly ARIN fees for 86 years.)
One difference between the ARIN region and the rest of the world is that in North America, a large amount of IPv4 address space was given out before the RIR system was created. Many of these "legacy allocations" are from before 1993. Back then, address blocks came in only three sizes: class A (16.8 million), class B (65,536), and class C (256). Many universities got class B blocks. A good number of large businesses and parts of the US and UK governments even received class As—which they've understandably been reluctant to give back.
These days, many systems share an address using Network Address Translation (NAT) and usable blocks can be as small as 8 or 16 addresses (although you need a block of at least 256 addresses if you want to trade). So holders of legacy address space have a lot of material to trade with. The holders of this space also never signed the contract with ARIN that spells out that "address space [is] not to be considered to be property."
So although the RIRs give out addresses to organizations that "need" them, the time horizon for the need justification has become shorter as the depletion of the IPv4 address space drew closer. This means that, for instance, Amazon would be able to get addresses for its Elastic Compute Cloud (EC2) service that are going to be deployed within a year. If it then needs more, it has to come back. By buying a large block of addresses, an organization can avoid the need to come back to ARIN for more addresses on a regular basis (and avoid the built-in unhappy ending).
ARIN maintains that transfers can only happen if the recipient can "demonstrate need," and in fact about a third of attempted transfers end up not happening. However, the researchers write that when Microsoft obtained Nortel's IP addresses in bankruptcy proceedings, the transfer policy was "bypassed."
"A last-minute intervention by ARIN, and private appeals to Microsoft, led to a compromise solution. Microsoft agreed to sign a special contract for legacy holders, known as a LRSA, and ARIN agreed that the transaction gave Microsoft the same de facto property rights held by the prior legacy holder, Nortel."
However, ARIN doesn't see it that way: "It is unfortunate that the paper includes conclusory material based on Milton Mueller's misconceptions of the first bankruptcy transfer," John Curran, president and CEO of ARIN, told Ars. Mueller is the lead author of the paper, and the bankruptcy transfer was where Microsoft bought 666 thousand addresses—I swear I'm not making this up—from bankrupt Canadian network equipment maker Nortel. Curran: "as it is clear that the specified transfer market is quite successful under the policies established by the ARIN community and this now includes numerous bankruptcy-related transfers which not only follow ARIN's policies but directly recognize ARIN's role in the transfer process."
When non-markets make sense
Although the IPv4 address market in North America is surprisingly robust since "free" addresses are still available, in the Asia-Pacific region, the numbers of addresses transferred don't amount to much. This holds especially true given the region's voracious appetite for fresh addresses in recent years. After using up more than 100 million addresses in both 2010 and 2011, it's hard to see how 0.7 million addresses transferred in six months can allow for business to continue as usual.
The reason I've always been very skeptical about address trading as a solution to the IPv4 address depletion is that I don't see how the world's largest ISPs, which use millions of addresses per year to connect new customers, are going to pay tens of millions a year to buy new addresses. Especially because it will get harder and harder to find sufficient amounts of tradable address space each year, pushing prices upward. Buying big Network Address Translation boxes and having users share an address is a more attractive proposition. As technology gets cheaper over time and as traffic moves to IPv6, fewer NATs can serve more users.
But we now know there are other players than ISPs looking to secure enough IPv4 addresses for the medium term. There's also the possibility that address trading will take off once trading between regions becomes a possibility, so that address-starved Asians can buy up addresses from North American companies such as HP. That company happens to be sitting on more than 33 million addresses. Or consider the US government, which has more than 168 million. Ultimately, maybe the money is better spent upgrading to IPv6 instead.
reader comments
70