AI and The Future of the Internet: Bigger
Dateline: October 26, 1997
This is part 2 of a 3-part series about the Internet, AI, and their mutual significance.
Each of the progressive developments outlined in the previous article in this series has contributed to making the Net a better environment, and this in turn has spurred growth. Now we turn to examine that growth, and some of its attendant problems.
Growth of the Internet can best be depicted graphically. A picture being worth a thousand words, an article on Internet growth can thus be kept short. There are many variables that can be used to describe growth, such as number of users and number of bytes transmitted over the Net. In the case of users, however, the data are unreliable. Bytes transmitted is reliably recorded, but is not a good indicator of general growth since it can be skewed by data-intensive transmissions, such as video or push services such as Pointcast.
The following charts show growth in the number of hosts (computers providing an Internet service), and the growth in bandwidth. In my opinion, these are among the more reliable and useful indicators of the general growth trend.
The data for these charts were obtained from ftp://nic.merit.edu/nsfnet/statistics/history.hosts and from Vinton Cerf’s September 1997 paper Beyond the Millenium: The Internet, available at http://www.mci.com/mcisearch/aboutyou/interests/technology/ontech/cerfreport0897.shtml.
Cerf’s 1997 paper mentions that 2 terabits (trillion bits) per second of bandwidth has been achieved in lab tests by running 128 colors of light through an optical fiber.
Problems
Three major problems confront the Internet: not enough addresses, routing tables growing too large, and network security. The first two (addresses and routing tables) are a direct result of growth, so we consider them below. Network security will be discussed next week.
Addresses
Every device—every computer, router, modem, telesensor and telerobot—attached to the Internet needs its own unique address in order for it to be accessible by other devices and humans on the Net. The addresses are known as IP (Internet Protocol) addresses, and are made up of 12-digit numbers in the form nnn.nnn.nnn.nnn. Addresses that use words and letters, such as www.mikiko.net, are for human convenience only—our machines translate such addresses to their underlying IP address before they are sent out over the Net. This is exactly how postal codes work. In the United States, every house has a unique10-digit "ZIP" code and in theory at least you can send a letter to the ZIP code without writing the actual street address, city, or state on the envelope.
There is a finite number of IP addresses currently. If nothing is done to increase the number, there will come a time when no more computers, routers, or other network devices can be added to the Internet. In 1996, this point was predicted to be reached in about 2004. Minor improvements to the IP address system occasionally add a little more breathing room, and as of 1997 Vinton Cerf was predicting "something like 2008" as crunch time, if nothing changed.
The current implementation of the Internet Protocol, known as IPv4 (version 4), has a theoretical limit of 4.3 billion addresses, which sounds like an awful lot. However, only five percent of the allocated address space is used, due mainly to an inefficient system of allocating addresses in large blocks to organizations that want to attach machines to the Net. Even if you only have one machine, you are given at least 256 addresses. This system effectively ensures that most IP numbers go to waste. It is as if governments issued only $100 banknotes: if you want to buy something costing $25, you have no choice but to pay $100, and $75 is wasted—there is no small change.
In September 1995, an Internet governing organization (they really should be called guiding organizations, since they have no formal power) approved four essential parts of the "next generation" protocol, appropriately named IPng. In theory, if every single number could be utilized (i.e., there would be 100 percent efficiency), IPng would supply 340,282,366,920,938,463,463,374,607,431,768,211,456 addresses, which sounds rather a lot more than 4.3 billion. And it certainly is. It works out to 665,570,793,348,866,943,898,599 addresses per square meter of the surface of the planet, which, you are no doubt aware, has 511,263,971,197,990 square meters.
In practice, 100 percent efficiency cannot be achieved, but even the most pessimistic IPng efficiency estimate forecasts 1,564 addresses for each square meter of the surface of the planet. The most optimistic estimate would allow for 3,911,873,538,269,506,102 addresses per square meter.
It won’t be easy to implement IPng, but it can and must be done.
Routing Tables
The second major Internet problem is the routing table problem. A routing table is a big table stored in memory in a router. It acts like a sorter in a post office, reading the addresses on all packets and sending packets out in different directions based on their IP addresses. Some organizations (for example, large universities) have thousands of individual computers, routers, and other devices on the network, while the Acme Automatic Cat Litter Box Corporation will have a mere handful. Yet the university and Acme "sorting boxes" (space in the routing tables stored in their Internet routers) are the same size. That is inefficient, and is becoming a problem as the routing table space in routers gets more congested. Just as there’s a finite limit on how many sorting boxes can be accommodated in a post office without moving the post office to a bigger building, so is there a limit on how big a routing table can be before a bigger router is needed.
Like building bigger post offices, changing to bigger routers is expensive. Worse, there seems to be a physical limit to the capacity and speed of routers, and that limit has almost been reached. In 1996 US Sprint, which own Sprintlink (one of the major Internet backbones) without warning decided not to accept packets addressed to certain "inefficient" sites. The sites affected were small ISPs that were not connected directly to Sprintlink (i.e., were not Sprintlink customers).
This issue, like the IP address space issue, poses a threat to the future development of the entire Internet. For that reason alone, we can expect strong efforts to solve it. Among several promising solutions already being worked on is one that would move the Internet away from a routed architecture and towards a switched architecture.
Next week, we will consider how adding intelligence to the Net is helping to counter these and other problems brought on by growth.
Until
next week,
NEXT WEEK: The Future of the Internet: Smarter.