IPv6 Transition to Cost US $75 Billion?

darthcamaro writes "There are alot of reasons why the US isn't moving as quickly as Japan and Europe in migrating to IPv6. One of those reasons is likely cost. An article on Internetnews.com cites an unreleased 'Dept. of Commerce report estimating it will take $25-$75 billion to pay for the transition.'"

Comments

Yee-Ha!

[ackthpt]

A mini-tech boom! Cisco will profit an anyone who makes switches which allow your old IPv4 stuff to communicate will make a fortune.

i'm applying for a patent on decaffeinated, low-fat, sodium free, left-handed wholly organic ipv6 veeblefetzers, axolotls and potrzebies

What is the basis for the cost?

[dotslashdot]

The FA makes no mention of WHY it will cost that much. I don't know anything about IP6, but $75b makes it seem like they plan on rewiring the whole government. The article cites that "one speaker" estimated the cost between $25-$75b. Is the speaker trying to just jack up the price? Perhaps someone can explain what is involved so we can decide if the prices quoted are reasonable.

Does anyone actually use english measures anymore?

[coyote-san]

Not to be dense, but does any manufacturer in the US still use english measurements? (Not consumer-facing products or places where legacy items are measured in english units.)

I believe car manufacturers switched to metric components years ago.

I'm sure aircraft manufacturers are also metric.

Consumer electronics? Considering that the last domestic manufacturer closed years ago I think it's a safe bet that it's entirely metric now.

Other industries?

Re:$25-$75 billion

[SatanicPuppy]

I don't know, I mean, where I work (and we have a metric ton of network hardware) transitioning to IP6, at least as far as dealing with the rest of the world, would be pretty easy.

As far as the internal network goes it'd be a nightmare, but in that case, why switch internally at all? No real need to at this point, we could do the translation without too much trouble. Let the internal stay IP4 until all the software/hardware becomes ip6 compatible, THEN switch.

Numbers like this are always pulled out of thin air. Sure it'd be a pain in the ass if we had to up and switch today, but it wouldn't be that bad to switch in 5 years or so if we mandated compatibility today.

Re:What's needed?

[anticypher]

The original report was by Juniper and presented to a group working on upgrading the U.S. government and military networks to be dual-stacked for both v4 and v6. Since Juniper sells very expensive equipment, they want to lessen the sticker shock for all their government buyers.

There are a lot (two words) of places to look for IPv6 dual stacking.

Start with the big IPv6 hardware equipment vendors, like Cisco, Juniper, and Foundry. Look at the (relatively) free implementations that exist today, like BSDs, OpenBGPd, Mac OS-X, some linux distributions, Windoze with a patch (and soon to be included by default in Vista). That will give you some background in what to do, but since you asked such a wide open question there isn't really any one place to point you. Its almost as if you asked "I need to set up the internet, is there someplace I can learn everything about it?"

Try subscribing to some IPv6 mailing lists, or at least browsing their archives. Lots to learn there, some technical, much political. Most of the political is from clueless noobs who have just barely caught on how to configure their home NAT router, and are terrified they will now have to spend another decade learning something slightly new. The real engineers consider the migration to a dual-stacked internet as just another excercise they have to do as with every new technology.

I will admit, there is a learning curve. I have over 20 years of IPv4 experience, and it still took me a while to pick up some of the subtleties of v6. BGP peerings takes some extra work, but then again, it took years to learn all I know about v4 BGP peerings.

I would love to see some of the major internet sites start serving up content via IPv6. Slashdot, which, unfortunately, no longer seems to have anyone technically competent running it, would be a huge boost to IPv6 if they started serving up AAAA records in DNS. Add extra karma during the first few months of early adopters who can connect with IPv6, and there would be a rush of competent geeks setting up IPv6 tunnels to their home networks and pressuring their upstream ISPs to support it natively.

There is a huge amount of work to be done before the internet can be dual stacked. Apache2 supports IPv6 addresses, but PHP, MySQL, Perl and a host of other apps/languages/scripts choke or die when presented with IPv6. The IETF working group moved IPv6 from draft to standard recently, and now we just have to wait until it works its way into more and more new devices. I'm waiting on Cisco to include IPv6 standard in all versions of IOS, just like IPv4 is now.

the AC

Re:$25-$75 billion

[rm69990]

As a Canadian, this rule doesn't fit. We have more land mass than the US, 10% of the population, cheaper/faster internet, and it is more reliable. For instance, where I live, the slowest Cable/DSL I can get is 3 MB/s, and it has gone down maybe 4 or 5 times in 2 years. The company sent out a technician, as we were the only ones in my area experiencing the problems, and it was faulty wiring in the house causing the problem. Since that was fixed, it hasn't gone down once. For $80 a month, we get phone with unlimited long distance (in the US and Canada), cable tv with a lot of extra channels and internet.

Asian Homes will use up IPv4, cellphones would

[billstewart]

As Asians get broadband at home, they're going to start burning a lot of IPv4 addresses, and to the extent that they deploy cellphones as IPv4 instead of IPv6, they'll also burn up lots of addresses. China's the main fast-growing player right now, but India may gradually get its act together and catch up. Japan and Korea are already heavily wired and/or wirelessed (insert canonical old-people-in-Korea joke here).

But there are hard problems and easy problems

• Linux, BSD, MacOS, and newer Windows versions all support IPv6 natively, though some of the support is rough (e.g. if your default DNS server doesn't know IPv6, things fail strangely or work slowly.)
• BIND and Apache support IPv6 if you want them to, and Firefox mostly does. So do some LDAP servers.
• Cisco, Juniper, and other major router vendors support IPv6 on newer equipment, but often their IPv4 support is in fast ASICs while the IPv6 uses slower CPU-driven processing so most ISPs haven't turned it on on most equipment, and IPv6 does take a lot more memory for tables because the addresses are bigger, and it seems that running dual-stack slows down everything compared to running IPv6-only and IPv4-only.
• Dialup Internet support comes from a really wide variety of equipment - some stuff can be easily upgraded, other stuff is antique. Fortunately, most dial users are clients who can use NAT or tunnel servers so they can ignore the problem.
• Home firewall routers like you've probably got under your desk don't support IPv6 - if you've got a Linux-based model, it might be upgradeable, but it might not have the horsepower to keep up with a fast broadband connection.
• Lots of NAT, proxy, and tunnel variants, 6to4 gateways, etc. exist, and can be used to hide non-IPv6 equipment from the scary new IPv6 network, or to let IPv6 equipment pretend to be IPv4 when talking to older networks, so there's a lot of transition support possible for people who want to use it - but it's potentially a lot of work.
• Lots of custom applications were written for IPv4 only, and would have to be rewritten to support IPv6 - it's a level of pain similar to getting Y2K bugs fixed. Some of them can be worked around by hiding behind 6to4 gateways, some can't. Sometimes it's just a database schema change, sometimes it's not.
• ISP network management databases often just know IPv4 - supporting IPv6 not only means changing out lots of tools, but also input screens for users, etc. Some ISPs are better at that than others, but as with equipment replacement, major software changes require money, and most ISPs aren't making much.

Re:Where's this cost coming from?

[Burdell]

OS and software updates are easy; people updates operating systems and other software all the time.

Infrastructure updates are hard. Routers last a long time. Cisco's dependence on CEF (Cisco Express Forwarding, aka Customer Enragement Feature) and hardware forwarding means that routers that can forward tons of IPv4 traffic can't handle a little IPv6 traffic (for example, the widely used 7500 series). Telling the boss that you need to spend $300,000 to replace one router (that oh by the way works just fine except for a feature nobody is asking for) doesn't go over well, especially when you have more than one router.

One of the widest used dialup concentrators is the Ascend/Lucent MAX and MAX TNT series. I believe UUNet used to use these for example (I don't know what they use now but I haven't heard of them changing); a lot of "national" ISPs resold UUNet dialup ports. TNTs have no IPv6 support at all even in the latest software updates (again, IIRC it is a hardware limitation). A lot of people still use dialup, especially when on the road; it is shrinking, so it is extra hard to spend big $$ replacing hardware that is operating just fine, but it isn't going to go away any time soon.

I work for a relatively small ISP, but we'd have to spend millions of dollars to support IPv6 across our network. AFAIK no customers are asking for IPv6; one friend asked informally if we had any plans and I said no and he went on to other questions.

Re:What's needed?

[Danathar]

This is ONLY true if blocks of address space are not dolled out in IPv4 fashion. The problem is that in the government and commercial world multi-homing to several ISP's for redundancy is the norm in IPv4. In an IPv6 envrionment there STILL is not a workable solution to having just about everybody subnetted.

I predict (and serveral people involved in IPv6 deployment on Internet2) that we'll end up giving /32's to MANY organizations because they'll want to connect to more than one ISP. Unless somebody comes up with a reasonable way for an organization with a /48 to be connected to two different ISP's (like my agency is under v4) for reduncancy.

Cost of transition

[netrangerrr]

The cost estimate we (Army CERDEC IPv6 Team) have done for the Army IPv6 transition leads us to believe essential $0 acqusitions costs if all IPv6 transition is done within regular tech refresh cycles. If we're buying IT gear anyway, IPv6 comes as regular product improvements over the next 3-5 years. The money DoD is spending at this point is aimed at getting MORE CAPABLE networks and at operations costs to train admins to run two IP stacks (v4 and v6) until we can phase out v4. By more capable, we are referring to new IPv6-only services like network mobility (NEMO) and multihoming (SHIM6).

You know...

[vertinox]

I hope that $75 billion includes fiber to the curb in every house in America.

I love IPv6 and all, but lets do the fiber first and then deal with the protocol.

Re:What's needed?

[Feyr]

everytime there's a discussion about ipv6 i bring up this point, and i get people like you that didn't read the policy giving the exact same answer.

see http://www.arin.net/policy/nrpm.html section ipv6 6.5.1.1

To qualify for an initial allocation of IPv6 address space, an organization must:
a) be an LIR; --- most ISP aren't
b) not be an end site; --- large hosting company ? i'm sure they'll appreciate having to renumber
c) plan to provide IPv6 connectivity to organizations to which it will assign /48s, by advertising that connectivity through its single aggregated address allocation; and
d) be an existing, known ISP in the ARIN region or have a plan for making at least 200 /48 assignments to other organizations within five years. ---- yeah right, 200 /48 ? so that's what, 50-100k customers? depending on your business model

Re:That's ridiculous

[evilviper]

And we don't have to wait for our ISPs, either. I've been using 6to4 (IPv6 tunneled over IPv4) for years.

Yeah, I'm sure 6to4 is going to work perfectly for everybody, particularly the US government. Who needs to buy new routers, when you can just tunnel everything? Woohoo!

Re:What's needed?

[Tony Hoyle]

You nicely stepped over the complexity there.

You had to custom modify a WRT54G with a working ipv6 stack and radvd, then sign up with a tunnel broker (precious few of these left now - most of the ones from a few years ago died), and manually edit scripts to connect to that tunnel broker.

Or you could have tried to go the 192.88.99.1 route, only to find that most ISPs don't route it any more.

Then you've got an ipv6 connection. woo. With a probable ~300ms first hop and nowhere to go because there is *zero* commercial deployment. Enjoy your address space.

Scalable addresssing for multihoming is just hard

[billstewart]

I've been trying to think about the multihoming problem for maybe five years, and the answers don't seem to be getting any better (:-) IPv6's advocates made a big deal in the early days about how they were going to give us magical solutions for scalable addressing and routing, and we weren't going to have all the end users in another big swamp with all the anarchy of IPv4 address assignment and just longer addresses, but the main thing that seems to have been done about it is have ICANN price IPv6 space high enough to discourage casual experimenters from getting their own space.

Some of the original motivations for end-users getting their own address space have gone away - DHCP means the cost of readdressing computers has gone to nearly zero, especially for desktop machines, and DNS means that you really _can_ move a server, though you might need a week or two of overlap on the address space for everybody's caches to time out (and IPv6 might force you to do some kind of tunneling deal with your old ISP), and firewalls mean that you might not by exposing most of your IP devices to the outside world anyway, just your public servers.

But even that doesn't mean that routing becomes much simpler, because that's only useful if you can aggregate - for instance, you could get /48 of provider-assigned address space from ISP1, and advertise that space on your connection to ISP2, so the global routing tables still have two entries for you even though they're assigned somewhat more cleanly. And aggregation doesn't always work well for geographically dispersed end-user customers - it's one thing for a university to aggregate the addresses from a bunch of buildings that are all in the same city, but if you've got a retail store chain with a thousand stores, should they all be part of one /48 at headquarters and route their external traffic there through IPSEC tunnels, or should each branch get provider-assigned address space from whatever ISP is nearby and try to tie that mess together, or some hybrid of both? For performance reasons, especially with VOIP, you'd like to keep latency down, so it's better to keep traffic in the same half-continent or so if you can, but it's not clear that there's an obvious answer.

IPv6 was supposed to free us from the evils of NAT. But the easiest ways to do multihoming either get into the swamp scalability problems or else do some kind of NAT or tunnel things to let you advertise one set of address space from two ISPs. Maybe that's not such a huge problem?