On Thu, 15 Jan 2004, Iljitsch van Beijnum wrote: > On 15-jan-04, at 18:12, Dean Anderson wrote: > > > But whether you internetwork with IPv6 and NAT, or just keep IPv4, NAT > > will not go likely go away. > > Directly internetworking IPv4 and IPv6 (where an IPv4-only host talks > to an IPv6-only host) is only possible for IPv6 hosts that use an > IPv4-compatible address. This is an interesting experiment, but it's > not very useful as this imports all the IPv4 address-related issues > into IPv6. I was thinking more of a dual homed host, with an IPv6 interface, and an IPv4 interface, and an IPv6-IPv4 NAT. Essentially your IPv6 HTTP request is NATTed or proxied through to an IPv4 server. > > The "math" below works out because 9 billion > > people don't each need a unique IP address. The vast majority of those > > people will be serviced via NAT, > > What is your basis for this argument? The current host count is around > 250 million but there are 2 billion addresses in (some form of) use. > This is obviously going to get better and in the long run not every > individual needs a public address 24/7, but I don't see how we can get > from 5 - 10 addresses per person to 0.3 without significant problems. Just the observation that most cable and DSL providers can give customers RFC1918 address space, and use NAT. Some providers are doing this now. The residential users don't need to have a globaly unique IP address. The vast majority of the 9 billion people in the world will be residential users. It could very well be possible for a long time to come that only a few billion unique addressses are needed. > > as cable and dsl providers are starting to do now. > > I don't know which part of the world you hang out, but I've never seen > any ISP NAT on behalf of its customers. I've heard some talk of GPRS > service providers doing this (but not seen it) but certainly not > broadband or regular dial-up. If there is NAT somewhere, it's done by > the end-users themselves. I've heard of residential service companies doing this. > > The other problem is that the entire world can't be converted at once, > > so there is significant business in the IPv4 side, that the IPv6 side > > can't abandon or lose connectivity to. > > So? One way to accomplish this is to install IPv6 everywhere and only > keep a few gateways around for access to "legacy" IPv4 web, email and > the like. Right, but everyone operating a server will want to keep their IPv4 legacy stuff, and what motivation do they have for going to IPv6? Why, for example, would Fleet Bank move to IPv6? This is a serious rhetorical question. You have to put yourself in the position of CIO of Fleet Bank, and consider the options. Internal cost, ISP cost, to what benefit? > > It would be like losing the ability to contact your bank and make > > deposits if you upgrade to the new version of the accounting software. > > You can't do that. And the bank can't upgrade because then it would > > lose the people who haven't upgraded. > > Way back when I had to use special software to interact with my bank, I > used to get a disk with new software in the mail several times each > year. I guess switching to a web based system was worth it for them > because of the savings on postage alone. Yep. Those programs cost the bank quite a lot. Most banks didn't do them. Some that did them, canceled them. My bank had a dialup program that they canceled after a pilot. > > Thus, legacy software is frozen in place, until it is no longer > > useful. This is why most of the software written, is still written in > > Cobol. It is likely that Cobol, and IPv4 will still be in use in > > 2050. > > That's ridiculous. None of us used IPv4 46 years ago, there is no > telling what we'll be doing in another 46 years. 46 years ago was 1958. In 1954, the first B52 bomber flew. The last one made was delivered in 1962. I note that we are still flying the B52. It was only a few years ago that the Pentagon shut down the last Multics machine. I think somewhere around that then the IBM 360 series was introduced. That same architecture still runs in the latest IBM Z series. You are right that we can't forsee the future. But we can make some guesses. Mine is that people will still be using Cobol and IPv4. I would also hope that people aren't using Microsoft, but I think that is probably unlikely. > > I think IPv6 would get faster deployment if the cost structure of IPv6 > > address space were different (free, or $35/yr registration fee), and > > one tried to get a specific new industry to adopt it like a > > proprietary protocol that just isn't propietary. My company, Av8 > > Internet, doesn't have IPv6 address space because it costs too much > > money > > Talk to ARIN. Here in Europe there is no additional cost for ISPs to > get IPv6 address space. Good point. > > for a plaything, and for many years to come, it is just for play. > > Unless you are a hardware or software vendor and your customers want > IPv6. I have yet to have an Internet customer ask for IPv6 address space. But I am certain I have customers that have done IPv6 testing. I think if you make operating systems and network software, the questions are different. But the developers of operating systems don't need to talk to Fleet Bank over IPv6 to test their software, and they don't need IPv6 space from their ISP. > > You observation that IPv6 is more than just expanded address space is > > good, but when given the chance to create a new protocol, people put > > all kinds of improvements in. These improvements increase the > > complexity somewhat, but I'm not sure they change the overall problem > > very much. Given the cost of the change, one would kind of want some > > significant improvement besides address space. > > What would this cost be? Are Windows XP and MacOS Panther more > expensive with IPv6 support than without? (You don't need IPv6-capable > routers or switches in order to run IPv6.) And IPv6 is still IP, it's > not hard to learn. Here's a 30 second list: Administrative--you have to create systems to track and manage two pools of space. Help Desk--you have to train people to support IPv6 Help Desk--you spend more time spelling out IPv6 addresses over the phone. Operations--you have to support routing 2 essentially separate protocols Operations--you have to manage and debug IPv6 problems. If you are servicing residential users that pay $35/mo, you can't spend too much time on the phone. You can't spend too much time on network operations. And you can't spend very much on resource administration. These are low margin operations. They can't spend very much per user and stay in business. They have to limit what their customers can do, just to limit those costs. IPv4 isn't hard to learn either, but the "unforeseen" support costs of IPv4 have driven a number of residential providers out of business. We don't offer residential services. The story is quite different again in the commercial services world. I think the commercial companies like high tech companies are more likely to want to simply play with IPv6. But that isn't something they can pay for, and it would still cost me money to provide the service. So what's my interest? But of course, when I talk to Cisco, I still ask them about IPv6 support. Of course, I'm not willing to pay for it either. Luckily, they are doing it "for free". Most everyone wants stuff that _can_ do IPv6, but actually using that support is another issue altogether. > Now if people prefer to stay with IPv4. that's fine. But the people who > made IPv6 happen deserve some credit because if and when the time comes > that we really need IPv6, it will be there, which is a very comforting > thought. I'm not saying that they don't deserve credit. I'm just saying that there are many other practicalities to consider.