> The question I would ask is whether on-demand IPv4 makes sense. To my > way of thinking, it amounts to deploying a new kind of network. We are > asking people to deploy IPv6 in their IPv4 networks, or to deploy IPv6-only > networks. That requires some portion of the money and time they have > available for such issues. Deploying an IPv4-on-demand network is another > thing competing for those same resources - it doesn't create new resources > or reduce the matters pertaining to IPv6 deployment - it creates another > demand for the same resources. I don't see the point. > > I suspect that the network actually routes IPv4 no matter what; what is being > handed out on demand is an IPv4 address to an edge device. Hosts right now > get IPv4 (and IPv6) addresses when they don't need them so they can use > them when they do. They would need a different mode of operation, > perhaps triggered by the resolver noticing that an application wants to access > some name and the name only has an A record. In that mode of operation, > the host only asks for (DHCP) an IPv4 address when it needs it, and routing in > the network is to the granted address for the lifetime of the address. > > Do I believe we can describe and solve that? Yes. Do I think we can do it more > cheaply and simply than moving folks and their applications to IPv6, and > convince operators to change their operational practices accordingly? Not > even close. I think it is a diversion from IPv6 deployment.
I think "on-demand IPv4" would be rather easy with PPPoE. Many (telco) ISPs are still using PPPoE, and there's still a lot of equipment and routers that support it. The costliest part of PPPoE was help desk calls for people who forgot their login/password. ISP-managed CE routers (that do PPPoE) have mostly evolved to have this auto-configured by the ISP. But not retail CE routers. In the early days of the Internet, before the need for persistent IPv4 connectivity -- by applications using keep-alives or autonomously and frequently checking in with some server -- was prevalent, many ISPs that used PPPoE supplied CE routers that timed out the PPPoE connection (usually after about 20 minutes of no WAN-bound IPv4 traffic). They also over-subscribed their IPv4 address space. The delay in establishing the PPPoE connection on-demand (when a WAN-bound IPv4 packet came from the LAN) was small; and in the context of sunsetting IPv4, such a delay would probably be quite acceptable. Chatty applications put an end to IPv4 address over-subscription and PPPoE connections that timed out. So I think an on-demand solution already exists (complete with aging equipment, code, and operational knowledge). For those who want it. Of course, no on-demand solution will work as long as there are still a lot of IPv4-only autonomously chatty apps. Disclaimer: Please don't take this email as a statement that I'm recommending this, or that my employer has any such plans, or that I have emotional attachment to this idea, or that I have some sort of agenda regarding this. I'm providing this purely as technical information, in case people weren't aware of it. Barbara _______________________________________________ sunset4 mailing list sunset4@ietf.org https://www.ietf.org/mailman/listinfo/sunset4