michael.dillon@xxxxxx wrote: >> A /48 per 'site' is good, especially in the case of >> businesses, for home-usage though, most very likely a /56 >> will be more than enough. As such IMHO having 2 sizes, one >> business, one homeuser, would not be a bad compromise, >> otherwise the really large ISP's, eg the ones having multiple >> million customers, would need multiple million /48's and then >> the address space consumption suddenly really goes really >> fast. Having /56's there would slow that down a little bit. A >> /56 is still 256 /64's, and I have a hard time believing that >> most people even on lists such as ARIN ppml or the various >> IETF ones will ever configure that many subnets at home. >> > > I would still like to get to the bottom of this issue and understand > what things a /56 assignment size will break. I strongly suspect that > these things will not be of great importance to networks in the home, > but I would still like to know what they are and document the issues > clearly. > IMHO "home network connection" is a misnomer. I'd call it "commodity network connection". The size of the network that is assigned to a home ends up being the size of a network that you can get "off the shelf", for a fixed price, with minimal support, and using commodity off-the-shelf mass-marketed equipment. Such networks, and such equipment, are used for a lot more than just residences. Here's one scenario that concerns me. Right now multiple layers of NAT are not unusual. The first NAT connects to the incoming network connection, and additional layers of NAT are plugged into that one. Maybe they have a wireless base station that includes a NAT and they plug that base station into their Ethernet router/NAT. People do this because it is convenient to be able to connect a network at any point to the existing network, and because using a NAT means that they don't have to do anything special to make it work like configuring subnet prefixes - the new "router" plugs in to the network and starts working. In IPv6 we really want to make it possible to not use NAT. So we need a way for routers to allocate subnets automatically to routers within their own networks. If the initial router gets a /48 it can fairly safely divide that up into (say) 16 /52s, or maybe 256 /56s, and give one of those out to any router plugged into its network that asks for a network. Then if another router is plugged into to one of those subnets, it can give out /56s or /60s, and so on. The scheme doesn't support arbitrary levels of nesting, but 3 layers is probably "good enough" in practice. But if the initial assignment on a commodity network connection is /56, that doesn't leave enough room for 3 layers and maybe not even 2 layers. What I find ironic is that the encroachment on "user bits" seems to be driven by a desire by ISPs to reserve enough bits for pre-allocation of address blocks to routers that don't exist yet or don't have a customer yet, and this impairs the ability of end-users to do the same thing. And I suggest that in both cases this is a really useful thing to do, and what we need is to figure out a way to make it work well for both ISPs and end users. > Also, I strongly suspect that the IETF did not consider the situation of > in-home networks in great detail when they reached the conclusion of /48 > for all sites, because at that time, there were few, if any, companies > planning Internet deployments on the same scale as the phone system. I > suspect that we have grown things a bit faster than was expected. > Again, if IETF needs to reexamine the /48 or other design decisions, we should do that. Keith _______________________________________________ Ietf@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ietf