If the number of addresses would be the only concern it would be much easier to expand IPv4 adddresses to 16 bytes and use the rest of IPv4. There are many more things in IPv6 besides address size. IPv6 or whatever besides IPv4 is needed. My point is not what is better, it is about how to internetwork both of them and this where layering comes into place. The logic is simple: IP was designed to internetwork networks. Once there is more than one IP network, let's internetwork them in a proper way as well. Not more simple than that. Given that, pure IPv6 hosts do not need to use v4 and the other way round. I'm running IPv6 network with IPv6 DNS since two years ago. It works fine for me, however I want to see both v6 and v4 in action. Do not mind to add private v4 here but internetworked with public v4 and v6 in the same way as v4 to v6 - internetworked but not hacked with the NATs. This is my "math". -----Original Message----- From: Iljitsch van Beijnum [mailto:iljitsch@xxxxxxxxx] Sent: den 15 januari 2004 15:03 To: Yuri Ismailov (KI/EAB) Cc: 'Keith Moore'; IETF Discussion Subject: Re: dubious assumptions about IPv6 (was death of the Internet) On 15-jan-04, at 12:48, Yuri Ismailov (KI/EAB) wrote: > I share pretty much the views expressed on the page. I don't share all of them (9 billion people in 2050, 3.7 billion usable IPv4 addresses, show me some math that makes this work) but where on this page is there a point being made? It seems more like a fairly random collection of statements. > Small addition to that would be the fact that current "transition" > proposals are killing the possibility of transition as such. IPv6 is > strongly dependent on IPv4 through various "ipv4 compatible" > addresses, which are essentially private IPv6 addresses, i.e. not > globally routable. ??? Which kind of IPv4-related addresses would a host with native IPv6 connectivity and IPv6-enabled applications need? > BTW, not the only private v6 addresses in this sense. This basically > means that it will be nearly impossible or at least extremely hard to > wash out IPv4 from the implementation. Personally I think that > removing IPv4 can not be justified. Removing IPv4 from the code would be a peculiar idea to say the least. But running IPv6-only works for the most part. I tried this on the most recent version of MacOS and it's very easy to disable IPv4 for all interfaces except loopback. If you manually configure a nameserver that's reachable over IPv6 and use IPv6-capable applications, no problems whatsoever. People do give me weird looks when I say that I want to test running IPv6-only, though. > Instead of introducing two networks internetworked in a proper way, > there are two networks tightly coupled to each other. Besides all, > such approach nearly exclude the design of yet another network, which > can be naturally internetworked with existing networks. Are there many > believers that IPv6 is the last one and forever? Forever is a very long time. IPv6 is far from perfect, but none of the issues are such that replacing the protocol because of them makes sense. If we're going to see something new it will probably be a niche-only thing for a very long time. But since a simple lack of address space won't be an issue in IPv6 (of course this leaves the complex lack of address space) it is almost certain that it will be possible to translate back and forth between the new protocol and IPv6 in a way that's compatible with existing IPv6 implementations, so deployment won't be an issue. > The power of layering was heavily underestimated if not ignored. Isn't > it clear enough that we would be nowhere if bridging of link layers > would take over internetworking of networks. Talking about power, where are you going to find the power to run those 16 million+ entry CAMs at 10 Gbps+ without a network layer?