Tuesday morning began with an entertaining and important call to action by Geoff Huston, on why we may have left transition to IPv6 too late and the serious consequences for open Internet connectivity that could result. It was recognised as early as 1992 that the 32-bit address space of IPv4 was going to be insufficient, indeed in those days this was expected to happen about 2002. The problem was postponed by the change from classful to CIDR routing, which was so successful that at one time the supply of addresses was expected to last till 2050. However the widespread adoption of home broadband and mobile connectivity increased the rate of usage and IANA issued its last block of addresses last year.
The solution to the problem – the greatly increased address space offered by IPv6 – has been known and deployable for many years. Indeed it is estimated that at least 50% of end user devices and 50% of backbone providers could already use it. Unfortunately between user devices and the backbone is the “last mile” where support is much less common. As a result most estimates suggest that less than 0.5% of Internet traffic uses IPv6. Most plans for an orderly transition from IPv4 to IPv6 included a period (perhaps several years) of parallel running, where those systems that supported both protocols offered them both and traffic gradually migrated as more and more paths became possible through v6. Unfortunately that relies on new devices being able to obtain both a v4 and a v6 address so long as they were needed to ensure that they could reach those services that did not yet have full v6 connectivity; in most regions of the world that will soon become increasingly difficult.
Instead there seems a significant risk that ISPs will try to maximise the use of the few remaining IPv4 addresses by re-using them through technologies such as Network Address Translation (NAT) and Application Gateways. NAT is already common at the edge of the network, in home routers and mobile connectivity, and can already limit what protocols can easily be used. UDP suffers particularly badly. Moving NAT further into the core of the network, with the likelihood that a given path will have to cross multiple NAT steps, seems likely to increase these problems. Application gateways cause even more problems because only those applications the gateway has been programmed to recognise can be used at all. Thus both developments are likely to reduce the Internet’s ability to support the development of new applications and services, since it will no longer be possible to achieve widespread use without persuading ISPs to support them in their application gateways.
This suddenly starts to sound like the network neutrality debate that has been taking place among European legislators and telecoms regulators, except that pressures to create a non-neutral internet now come from technological issues, rather than from economic ones. Perhaps the political interest that has emerged in encouraging ISPs to consider the open Internet as more important than short term financial interests might also be used to encourage them to adopt a long-term, rather than short-term, approach to the IPv6 transition?