by Geoff Huston, APNIC
At the recent APNIC meeting in New Delhi, the subject of IPv4, IPv6, and transition mechanisms was highlighted in the plenary session . This article briefly summarizes that session and the underlying parameters in IPv4 address depletion and the transition to IPv6.
As of September 2007 we have some 18 percent of the unallocated IPv4 address pool remaining with the Internet Assigned Numbers Authority (IANA), and 68 percent has already been allocated to the Regional Internet Registries (RIRs) and through the RIRs to Internet Service Providers (ISPs) and end users. The remaining 14 percent of the IPv4 address space is reserved for private use, multicast, and special purposes. Another way of looking at this situation is that we have exhausted four-fifths of the unallocated address pool in IPv4, and one-fifth remains for future use. It has taken more than two decades of Internet growth to expend this initial four-fifths of the address space, so why shouldn't it take a further decade to consume what remains?
At this point the various predictive models come into play, because the history of the Internet has not been a uniformly steady model. The Internet began in the 1980s very quietly; the first round of explosive growth in demand was in the early 1990s as the Internet was adopted by the academic and research sector. At the time, the address architecture used a model where class A networks (or a /8) were extremely large, the class B networks (/16) were also too large, and the class C networks (/24) were too small for most campuses. The general use of class B address blocks was an uncomfortable compromise between consuming too much address space and consuming too many routing slots through address fragmentation. The subsequent shift to a classless address architecture in the early 1990s significantly reduced the levels of IPv4 address consumption for the next decade. However, over the past five years the demand levels for addresses have been accelerating again. Extensive mass-market broadband deployment, the demand for public non-Network Address Translation (NAT) addresses for applications such as Voice over IP (VoIP), and continuing real cost reductions in technology that has now brought the Internet to large populations in developing economies all contribute to an accelerating IPv4 address consumption rate.
Various approaches to modeling this address consumption predict that the IANA unallocated address pool will be fully depleted sometime in 2010 or 2011 [2, 3, 4, 5].
Transitioning to IPv6
The obvious question is "What then?", and the commonly assumed answer to that question is one that the Internet Engineering Task Force (IETF) started developing almost 15 years ago, namely a shift to use a new version of the Internet Protocol: what we now know as IP Version 6, or IPv6. But if IPv6 really is the answer to this problem of IPv4 unallocated address-pool depletion, then we appear to be leaving the transition process quite late. The uptake of IPv6 in the public Internet remains extremely small as compared to IPv4 . If we really have to have IPv6 universally deployed by the time we fully exhaust the unallocated IPv4 address pools, then this objective appears to be unattainable during the 24 months we have to complete this work. The more likely scenario we face is that we will not have IPv6 fully deployed in the remaining time, implying a need to be more inventive about IPv4 in the coming years, as well as inspecting more closely the reason why IPv6 has failed to excite much reaction on the part of the industry to date.
We need to consider both IPv4 and IPv6 when looking at these problems with transition because of an underlying limitation in technology: IPv6 is not "backward-compatible" with IPv4. An IPv6 host cannot directly communicate with an IPv4 host. The IETF worked on ways to achieve this through intermediaries, such as a protocol to translate NATs , but this approach has recently been declared "historic" because of technical and operational difficulties . That decision leaves few alternatives. If a host wants to talk to the IPv4 world, it cannot rely on clever protocol translating intermediaries somewhere, and it needs to have a local IPv4 protocol stack, a local IPv4 address, and a local IPv4 network and IPv4 transit. And to speak to IPv6 hosts, IPv6 has the same set of prerequisites as IPv4. This approach to transition through replication of the entire network protocol infrastructure is termed "Dual Stack." The corollary of Dual Stack is continued demand for IPv4 addresses to address the entire Internet for as long as this transition takes. The apparent contradiction here is that we do not appear to have sufficient IPv4 addresses in the unallocated address pools to sustain this Dual Stack approach to transition for the extended time periods that we anticipate this process to take.
What Can We Expect?
So we can expect that IPv4 addresses will continue to be in demand well beyond any anticipated date of exhaustion of the unallocated address pool, because in the Dual Stack transition environment all new and expanding network deployments need IPv4 service access and addresses. But the address distribution process will no longer be directly managed through address allocation policies after the allocation pool is exhausted.
Ideas that have been aired in address policy forums include encouraging NAT deployment in IPv4, expanding the private use of IPv4 address space to include the last remaining "reserved-for-future-use" address block, various policies relating to rationing the remaining IPv4 address space, increased efforts of address reclamation, the recognition of address transfers, and the use of markets to support address distribution.
Of course the questions here are about how long we need to continue to rely on IPv4, how such new forms of address distribution would affect existing notions of fairness and efficiency of use, and whether this effect would imply escalation of cost or some large-scale effect on the routing system.
On the other hand, is IPv6 really ready to assume the role of the underpinning of the global Internet? One view is that although the transition to a universal deployment of IPv6 is inevitable, numerous immediate concerns have impeded IPv6 adoption, including the lack of backward compatibility and the absence of simple, useful, and scalable translation or transition mechanisms . So far the business case for IPv6 has not been compelling, and it appears to be far easier for ISPs and their customers to continue along the path of IPv4 and NATs.
When we contemplate this transition, we also need to be mindful of what we need to preserve across this transition, including the functions and integrity of the Internet as a service platform, the functions of existing applications, the viability of routing, the capability to sustain continued growth, and the integrity of the network infrastructure.
It appears that what could be useful right now is clear and coherent information about the situation and current choices, and analyzing the implications of various options. When looking at such concerns of significant change, we need to appreciate both the limitations and the strengths of the Internet as a global deregulated industry and we need, above all else, to preserve a single coherent networked outcome. Perhaps this topic is far broader than purely technical, and when we examine it from a perspective that embraces economic considerations, business imperatives, and public policy objectives, we need to understand the broader context in which these processes of change are progressing .
It is likely that some disruptive aspects of this transition will affect the entire industry, and this transition will probably be neither transparent nor costless.