We don't need subnets any more

In his regular monthly spot on PING, APNIC’s Chief Scientist Geoff Huston discusses the question of subnet structure, looking into the APNIC Labs measurement data which collects around 8 million discrete IPv6 addresses per day, worldwide. Subnets are a concept which "came along for the ride" in the birth of Internet Protocol, and were baked into the address distribution model as the class-A, class-B and class-C subnet models (there are also class-D and class-E addresses we don't talk about much). The idea of a sub-net is distinct from a routing network, many pre-Internet models of networking had some kind of public-local split, but the idea of more than one level of structure in what is "local" had to emerge when more complex network designs and protocols came into being. Subnets are the idea of structure inside the addressing plan, and imply logical and often physical separation of hosts, and structural dependency on routing. There can be subnets inside subnets, its "turtles all the way down" in networks. IP had an ability out-of-the-box to permit subnets to be defined, and when we moved beyond the classful model into classless inter-domain routing or CIDR, the idea of prefix/length models of networks came to life. But IPv6 is different, and the assumption we are heading to a net-subnet-host model of networks may not be applicable in IPv6, or in the modern world of high speed complex silicon for routing and switching. Geoff discusses an approach to modelling how network assignments are being used in deployment, which was raised by Nathan Ward in a recent NZNOG meeting. Geoff has been able to look into his huge collection of IPv6 addresses and see what's really going on.