Let’s take your first example: a 65536-pixel image. If each pixel represented 256 addresses, then the image would represent 65,536 (pixels) * 256 (address/pixel) = 16,777,216 addresses. Far from 4,294,967,296, the total number of IPv4 addresses…

In reality what you mean to explain is that each PIXEL SIDE represents 256 addresses, so each PIXEL represents 256*256 = 65,536 addresses.

Too bad there is no killer app yet, that would accelerate the migration.
cheers,
John

Mass of eath in kg = 5.9736 x 10 ^24
So in grames = 5.9736 x 10 ^27

2^128 / 5.9736 x 10 ^27 = 56 Billion

56 billion addresses per GRAM of matter on earth!

2^128=340282366920938463463374607431768211456 – total addresses
sqrt(2^128) = 2^64 = 18446744073709551616 – addresses along one edge (Area of a square = length * width)
105 pixels per linear inch * 12 inches per foot * 5280 feet per mile = 6652800 pixels per linear mile
2^64/6652800 ~= 2772778991358 address miles per pixel
256 addresses * 105 pixels per linear inch * 12 inches per foot * 5280 feet per mile = 1703116800 addresses per mile
2^64/1703116800 ~= 10831167934 linear miles
2^32 addresses * 105 pixels per linear inch * 12 inches per foot * 5280 feet per mile = 28573558426828800 addresses per mile
2^64/28573558426828800 ~= 645 linear miles

The math checks out fine.

A brief history.

IPv4 was developed in 1974-75 out of a need for a universal addressing system for global routing. Back then smart phones, wireless laptops, and even PCs did not exist. (The microprocessor chip had just been invented 2 years earlier, and the very first microcomputer kits, like the Altair 8080 were just coming into existence; but these were more “geek toys” than practical computing or networking devices.) Computers were big, monstrous devices that took up large rooms – or at least large closets – and cost tens- or hundreds of thousands of dollars. Only large corporations and universities had computers, and only a small percentage of their employees knew how to operate them. In that environment, 4.3 billion addresses seemed like a virtually unlimited supply. I’m not sure of the exact figure, but I’m guessing that there were less than 100,000 computers (or at most a few hundred thousand) in the entire world at that time. So the designers of IP were not concerned with address space efficiency and designed the classful system that only allowed for three sizes of networks; class C (254 hosts or smaller), class B (up to 65,534 hosts), and class A (up to 16,777,214 hosts). At the time, they anticipated that the vast majority of network address requests would be for class C. The classful addressing system allowed for 126 class A, 16,384 class B, and just over 2 million class C networks in total.

When IBM introduced the PC in 1981, they legitimized the microcomputer as both a business tool and as a networking device. With the PC boom came a significant increase in the rate of requests for class B networks. Organizations came to realize that with a PC on every desktop, networks capable of supporting only ~250 host devices would not be large enough to accommodate future growth. This is when the “IP engineers” started to worry about the limits of the IPv4 address space. It was not that they would run out of IP *addresses* per se, but rather that they would run out of IP *networks* – specifically class B networks – by the early 1990’s. Once the 16 thousand class B networks were allocated, we could start giving organizations class A networks, but with only about 100 of them, that would not last long.

The first significant “address stretching” solution introduced in the mid-1980’s was subnetting, and was later refined with Classless Inter-Domain Routing (CIDR) and Variable Length Subnet Masking (VLSM). This allowed organizations to divide larger class B and class A networks into smaller subnets and allocate only the portion of a classful network that an organization needed. Combined with re-claiming unused address space from previously-deployed class B and class A networks, this eased the address crunch.

In 1991-92, the World Wide Web was introduced. HTTP gave a user-friendly interface to the previously “geeky” Internet. This, combined with increased accessibility through service providers like CompuServe, AOL, and MSN, again accelerated the rate of IP address allocation. It was estimated that somewhere around the turn of the millennium we would run out of IP address space. Another pair of “stretching” solutions – private addressing and Network Address Translation (NAT) – eased the crisis. Using private IP addresses locally and NAT for global services, a local network could have hundreds or even thousands of devices sharing only a few or even one global IP address. Instead of end-organizations being separate networks on the global Internet, each was simply a node (or a few nodes) on the service provider’s network. The widespread adoption of private addressing and NAT forestalled IPv4 address exhaustion into the 21st century. Still, each new customer required at least one global IP address, and the number of global addresses was starting to run very low.

The most recent surge in IP address allocation came in the first decade of the new millennia. The popularity of mobile devices – wireless laptops and more recently smartphones – have again accelerated the consumption of IP addresses. On January 31 of this year (2011), IANA – the international “keeper of the IP addresses” – allocated the last blocks of IPv4 address space to APNIC, the Asia/Pacific regional Internet registry. As the regional registries allocate their remaining addresses to ISPs, they cannot replenish their stores. As the ISPs assign those remaining addresses to their customers, they, too will not be able to replenish. Once they’re out – that’s it. Nada. Nil. Over. Done.

Local networks can still use IPv4 for as long as they want, but the ISPs will have to switch over to IPv6 if they want to continue to grow and expand their customer base. Customers who choose to stick with IPv4 locally will eventually still have to accommodate IPv6 on their WAN connection to their ISP (just as Comcast cable TV customers eventually had to get digital TVs or converters as the cable company phased out analog service). Any telecomm service providers that don’t migrate to IPv6 will find themselves dying by virtue of being obsolete. Sooner or later, the end-user networks will also make the switch over. (Does anyone still use DOS or Windows 95? Yes, but they’re technological “hermits” who exist with no support or sympathy.)

The bottom line – while we’ve done a remarkable job in keeping this 37-year-old technology running this long (kind of like my grandfather’s 1967 Mustang), we are now at the point where it is beyond repair and it is time to replace it with a new, bigger, and better Internet Protocol.

p.s. As to Jorge @16’s question about IPv5, there was an experimental protocol (Internet Stream Protocol, or ST) developed in the late 1970’s that was dubbed “IP version 5” (ca. 1994). Its intention was to create a protocol that would better handle virtual circuits and streaming data. It was never widely deployed, and has since been replaced by better protocols like Asynchronous Transfer Mode (ATM). More to the point; ST (or IPv5, if you will) did not create a new addressing system – it used IPv4’s addressing scheme and therefore is/would be in the same predicament.

As Julien @14 and mamou @17 pointed out, there are a few errors in Aaron’s math.

For the first image, each pixel represents 2^16 = 65,536 IPv4 adresses, not 256. Keeping that pixel representation (65536, not 256) in paragraph 6 makes Aaron’s monitor’s pixel size correct, but his converting that to miles is off by a factor of 10 – it would need to be about 10.8 billion miles, not 1.08 billion. That’s approximately where the Voyager I spacecraft is now, 33 years after it was launched. (This is considered outside the sun’s “sphere of influence”, i.e. outside our solar system.)

In the next paragrph (P 7), if we allocate an entire IPv4 address space per pixel, we’d need a monitor that was 2^48 = 281,474,976,710,656 pixels on a side (Aaron’s value is off by a factor of 4096). That monitor, at Aaron’s pixel density woul still have to be ~42.3 million miles on a side – about the distance from Mercury to the sun.

For Aaron’s “six-state” monitor illustrated in the third image, each pixel would need to represent 2^64 (~18.4 quintillion) IP addresses, not the 2^32 (~4.3 billion) that Aaron suggests.

With these corrections, this still makes an awesome visualization of the vastness of IPv6 space.

More comments on other readers’ comments next…

If, from the 128-bit IPv6 addressing space, a block of addresses equivalent to the entire IPv4 addressing space (~4.3 billion addresses; an entire “Internet”) were assigned every *nanosecond* since the dawn of the universe -– not the dawn of mankind or the dawn of the Earth, but from the big bang itself -– we would up to now have assigned less than 1% of the available IPv6 space, and would have enough addresses still left to keep going for another 2.5 trillion years. (Note that the universe is “only” about 13.7 billion years old, yet we would have enough addresses left to keep going for another 2.5 trillion years.)

I think IPV4 will be around for a long time, since NAT provides a way to “hide” the real IPV4 addresses behind routers. Technologies often seem to live on well past their prime, simply due to the investments that have been made in them, and the costs of switching to a newer, better technology.

After all… we’re still using COBOL! I worked with a client last summer who has an entire claims processing system running in Cobol… on Unix, no less! It will likely still be running long after I retire… on IPV4! :O)