Doomed, doomed I say!
Posted by aogMonday, 14 July 2008 at 09:14 TrackBack Ping URL

In another lost thread, I read somewhere of the claim that the Internet is going to collapse around 2011 due to running out of IP addresses. It’s a problem, but hardly a crisis. The root of the problem is that when IP was first designed, they figured a few thousand computers was a huge number of computers to hook together, so they created a system with 4 billion addresses to cover that.

In practice, one can’t hook any where near 4 billion computers together via IP because of certain inefficiencies in the protocol and we are approaching the point at which adding computers will become very difficult. I suspect that had the original designers known of this problem they would have figured “it’ll be decades before then, surely better solutions will arise by then”. And they would have been right, because an updated version of IP (“IP version 6, or IPv6”) is in fact available. The address space is so large that every single person on Earth could be assigned as many addresses as there are in the current version (IPv4) without significantly impacting the available addresses. There are enough addresses to assign one to every individual atom on the entire surface of the Earth. That should last a while.

So what’s the problem? Well, as with any transition, there will be quite a bit of pain and expense to transition. Part of the problem is that it’s difficult for machines running IPv4 to talk to machines running IPv6, because the latter may have an address that the IPv4 machine can’t express.

Still, it’s just painful, not insurmountable. I remember having a conversation about this 5 years or so ago with my grand boss. He thought the ISPs would switch first. I told him no, because (at the time) the customers of the ISP would no longer able to be addressed by anyone else, and while that wouldn’t be a problem for most, it would be for enough that it wasn’t realistic. Instead, I expected it to show up in the ISP core first. It turns out that to make a connection between two routers, you need 4 IP addresses1. But those addresses are only used by the two routers involved, and nothing else on the planet. So converting such connections to IPv6 has zero impact on customers, meaning no nasty calls to tech support. My view was that the supply of addresses became tight, their value would increase, and ISPs would do such conversions in the core so they could re-sell the addresses to customers. And lo, that’s how it played out over the last few years (yes, I gloated terribly). The next step, IMHO, is the conversion of corporate campuses. If a company converts its entire physically adjacent network to IPv6, everything still works internally and the company can then either re-sell the addresses or stop paying for them. Externally connectivity can be handled via NAT2. Corporate conversions will likely carry us for a number of years beyond the 2011 deadline.

Beyond that, Vista is IPv6 ready, and so is the current version of Linux. In 5 or 10 years, when the corporate conversion well starts running dry, ISPs will start charging customers more for keeping an old style IPv4 address because 80+% of their customers will be able to switch over to IPv6. After that, webhosting companies (which will have long previously allocated IPv6 addresses in parallel to the IPv4 ones) will start charging, gradually forcing out IPv4. The tipping point will be reached somewhere in there and IPv4 will pass the way of Netscape 4.3 and dialup.

The only possible failure mode is heavy government intervention, which I certainly wouldn’t rule out.

1 If you don’t already know, you don’t want to.

2 NAT is the network equivalent of a forwarding address. Think of a corporate office that shares a single street address. The mail boy keeps track of which employee sent mail where so when a reply comes back he knows who it is for, even though the employee’s name isn’t on the envelope. The cost is that outsiders can’t initiate exchanges with employees, but a corporation is likely to see that as a feature, not a bug.

Comments — Formatting by Textile
Gronker Monday, 14 July 2008 at 19:07

As a former ISP owner and network designer, I see no need to move to IPv6. IPv4 has, largely, become a backbone addressing space. I saw a recent dump from a very large eCommerce company and over 97% of their customer browsers were on non-routable numbers. More and more ISPs are pushing non-routable numbers out to clients unless the pay for the “privilage” of a static, routable number. And most people dont even know what that is.

IPv6 certainly has enough address space to cover any forseeable need, but still routes slow on most stacks… and more importantly isnt really needed. Will it be someday? Its always possible, but I think you will see a true net of nets of IPv4 before that, parallel addressing spaces “NAT”ted between each other. But thats a post for another time.

Annoying Old Guy Monday, 14 July 2008 at 19:35

But what about the gamers? They’d notice if they had non-routables. I don’t know anyone with a non-routable for their ISP connection. Generally by “static” the ISP means “unchanging”, rather than being DHCPd.

I do know that a lot of the top end ISPs have switch to IPv6 in the core, because the routing in those cases is trivial (i.e., it’s a point to point connection with two hardwired addresses).

On the other hand, I suspect that if something does kill the Internet, it will be routing limitations more than IP address limits, even with IPv4. And on that score, it is possible that IPv6 would, overall, improve routing by reducing the fragmentation of address spaces.

Gronker Tuesday, 15 July 2008 at 01:45

Generally by “static” the ISP means “unchanging”, rather than being DHCPd.

It has in the past, but recently I have seen large ISPs handing out non-routables for everything but “static” (surcharged) connections. This is very likely to continue.

And the gaming front, most games deal with NAT just fine these days. Only the point to point games and protocols are still having issues. SIP still being a major pain in the ass. And that is improving as STUN/TURN/etc issues are being worked out.

I think the biggest issue with IPv6 is not really the cost, since as you mentioned, even the slow adopters are have come around and support some v6 stack. Its the fact that it will NOT “unfracture” the address space, but just the oppostite. We, as programmers, adminstrators, and humans, seem to expand our use of resources to the fit the available pool. When everyone and their dog has universally addressable (and therefore potentially routable entries) toasters and bathroom outlets - what then? Sure, IPv6 allows for a heirarchical structure, but eventually, everyone is going to want “number portablilty” for their toaster and the it all comes crashing down.

The first (and second and third and forth) IP number panics all lead to a further use of protocols like NAT to force organizations and households. This has been “a good thing”™.

God help me, but I think I need to appologize to AOL for their “World Wide Proxy” stance of the late 90’s… but they may have accidently stumbled on to something :)

Gideon7 Tuesday, 15 July 2008 at 02:25

A big problem is the lack of a widely adopted encapulsation/proxy system to facilitate migration. For example the only public Teredo Server that I know of in the United States is, and it doesn’t work half the time.

Annoying Old Guy Tuesday, 15 July 2008 at 08:07


Why not buy a proxy router? They’re less than $100 and rather widespread.


“Dealing with NAT” just means having a server with a routable address run by the game company. But there are still a lot of point to point or private server games out there. I don’t see that changing, as a lot of people don’t want to be in a truly public game or want to customize in ways that the game company sponsored servers won’t do.

I don’t see people wanting number portability. They’ll want name portability, but that’s much easier.

I suppose we’ll find out. But I think we agree that there is not going to be any real panic or collapse in the foreseeable future.

Gideon7 Tuesday, 15 July 2008 at 09:59

If my IPv6 computer/router wants to talk to your IPv6 network over the public IPv4 Internet there needs to be a broker to mediate the connecting tunnel. Right now there isn’t any public broker to speak of.

Annoying Old Guy Tuesday, 15 July 2008 at 11:43

Ah. Yes, that’s why I think ISP customers will be the last to switch over, and won’t do so until there’s enough network effect so that issue isn’t an issue. For a corporate campus, it’s not an issue because the corporation will put the necessary proxies on the perimeter.

On the other hand, if Gronker is right, the ISP will provide the proxy for its customers, because they must already be doing that if they’re using non-routables for customer IP addresses.

Post a comment