EXTRAPOLATION can lead you up strange streets, especially when those streets are in cyberspace. Last November Matthew Grey, a researcher at the Massachusetts Institute of Technology, set a piece of software loose on the Internet which tried to count how many separate 鈥減ages鈥� there were on the World Wide Web, the preferred system for publishing and retrieving information on the Internet, and how quickly new ones were being added.
The Web only began to be a functional part of the Internet in 1990, providing users with the first logical grouping of related information. Grey鈥檚 data found that, at that time, the Web was growing at 1 per cent per day; therefore, based on the number of users connected to date, every person in the world would have their own personal Web page within five years.
Yet in September, a separate estimate of the growth of the Internet, by Network Wizards, an American consultancy specialising in the subject, suggested that at present rates, everyone on the planet would be connected by 2003.
Advertisement
Clearly, there is a logical contradiction. How could everyone have a Web page, yet be three years away from being connected to the Internet? But perhaps there is another explanation: perhaps the rate of growth changed so much between September and November that it altered the Internet鈥檚 鈥渟aturation date鈥� 鈥� the time when, in order for it to expand, we will have to locate aliens in order to offer them e-mail addresses.
Both seem unlikely. But there is a third and far more likely explanation: the numbers are wrong. This is because nobody truly knows how many people are connected to the Internet. There are only estimates, and methods of estimating. But even those differ in approach and accuracy.
At the turn of this century, religious academics became greatly exercised over the question of how many angels could dance on the head of a pin. Arguing about how many people are connected to the Internet is the modern equivalent. The latest estimates range from 16 million to 40 million; not surprisingly, those in the business prefer numbers towards the top of that range.
To understand why it is so difficult to be definite, it is necessary to look at the structure of the Internet and how it has grown. One of the most notable features of the network is that it has no central authority. 鈥淚t鈥檚 nothing more than a bunch of computers connected together, to which a bunch of other computers dial in to, and through 鈥� just a big wide-area network,鈥� says Donald Taylor, a computer consultant from Hampton, Virginia. All that is needed is a standard means of transferring data between those computers, known as the Internet Protocol (IP), and you鈥檙e off.
However, the Internet is not totally anarchic. Every computer has to be able to locate other computers, in order to pass messages to them. To do this, some of the computers in the network act as reference points for other computers, to tell them where other machines are 鈥� in the jargon, what the machines鈥� IP address is. A machine that links directly to the Internet is known as a 鈥渉ost鈥�; host computers are arranged by categories 鈥� commercial, military, academic, or non-profit 鈥� known as 鈥渄omains鈥�. Anyone linking a new system directly to the Internet must register what they want their system to be called with the 鈥渄omain name server鈥� company (see Diagram).
In the US, this is a group called the Internet Society. In Britain, a company called Pipex looks after registration of commercial organisations, while the administrators of the Joint Academic Network (Janet) 鈥� one of the networks connected to the Internet 鈥� registers computers from academic organisations. Registering the system is an electronic christening.
On the Internet, a computer鈥檚 鈥渄omain鈥� is the last part of its electronic mail address. For example, 快猫短视频鈥檚 London offices have a machine whose e-mail address is edit@mail.newsci.ipc.co.uk. This machine鈥檚 鈥渄omain鈥� is the section.co.uk. Other e-mail addresses in Britain end with the letters .ac.uk (for academic institutions) and .org.uk (for other organisations, such as schools or non-profit groups).
If somebody in the US, for example, posts an e-mail to 快猫短视频, the message is passed to their local Internet link, which is the nearest computer with a direct link to the Internet. That machine examines the e-mail address from the right-hand side backwards to find the domain 鈥� from which it recognises that the company is registered in Britain. The US computer then sends a message over to Pipex鈥檚 DNS machine in London, which holds the list of all the machines on the Internet with the suffix.co.uk.
The Pipex machine confirms that the 鈥渉ost鈥� computer mail.newsci.ipc exists in the .co.uk domain, and sends back a string of numbers which constitute the IP address of the host computer mail.newsci.ipc.co.uk. The machine in the US attaches this string as the destination header for the e-mail message, and passes it on to a routing computer. The e-mail is then passed through the Internet until it reaches a computer whose IP address matches that on the header. All that one computer needs in order to be able to send e-mail to any other computer is the IP addresses of the various domain name server (DNS) systems because they hold the more detailed information needed to route the message.
With this kind of checking and double checking as an e-mail message is sent from one side of the world to another, it is tempting to think that the DNS machines, with their list of all the registered hosts, is an ideal source of data for estimating how many people are 鈥渙n the Net鈥�. But there are four problems with this.
The first is that there is no direct relationship between a host machine and the number of users. 快猫短视频鈥檚 London office has five computers with registered host names. But many more than five people have access to it, and all would legitimately say that through the system they are connected to the Internet. This situation is repeated all over the world.
Secondly, many Internet service providers offer Internet accounts for individuals, who gain access to them by dialling in to a central computer. But not everyone has a specific IP address; instead, it has a 鈥減ool鈥� of, say, 30 IP addresses. When someone dials in and there is a spare address, it is allocated to them. If more than 30 people try to access the central system at once, the latecomers will get a busy signal. However, once one of the first 30 users disconnects their modem, or closes their Internet application, an address is returned to the pool, and it can be allocated to someone else. This way, hundreds of people might claim to be connected, yet the DNS for that company only lists 30 IP addresses.
The situation is even more confused by the presence of companies like America OnLine, Prodigy and CompuServe, which offer a 鈥渂ulletin board鈥� service to millions of people worldwide. Their systems operate independently of the Internet, so that subscribers can find a range of free and paid-for services that are not available on the Net itself.
Until last year, they were effectively closed communities: although subscribers could swap e-mail with people over the Internet, they did not have access to more usual Internet features such as remote file transfer to or from Internet sites, newsgroups or the World Wide Web. Last September, these three American companies had a worldwide total of almost five million subscribers (see Diagram). Even later last year, all three began offering their subscribers Internet access, which meant that the number of people on the Net must have increased dramatically; but the host allocations would barely reflect that, because they use a 鈥渇loating address鈥� system.
Keeping accounts
Next, some people have more than one account, through different Internet connections. It is not unusual for people to have Internet access from their workplace and also a private account for home use. Only one person is actually connected, but there are two host allocations.
Finally, more and more companies are putting their host machines behind a 鈥渇irewall鈥� 鈥� a piece of software which, as a security measure, allows data out to the Internet, but severely restricts that which can come in. This prevents hackers breaking into the system, but has the side effect of making the host 鈥渋nvisible鈥� to the Internet at large. This is because another standard method of estimating the number of hosts, besides counting those registered at the DNS systems, is to use a program called 鈥減ing鈥�, which sends a data packet 鈥� like a brief e-mail 鈥� to a host computer, to see how long the host takes to acknowledge its arrival. If, however, a host is using a firewall, the data packet will not be allowed in, or acknowledged. Although properly addressed e-mail will reach users on that host, the machine appears not to respond to anonymous requests to identify itself (see Diagram).
So, the most quoted figures for the growth of the Internet will be for the number of hosts, because this is easily obtained from the size of the DNS machine鈥檚 database. When Network Wizards did this recently, it found that the last quarter of 1994 showed the largest growth in recent history 鈥� an annual rate of 26 per cent 鈥� pushing the numbers up to 4 851 000 hosts. But even that is an underestimate, says John Quarterman, of the Internet consultancy Matrix Information and Directory Services (MIDS), based in Austin, Texas, and widely regarded as an authority on how to measure the growth of the Net. 鈥淭hat takes a linear monthly snapshot and projects it as if the growth rate of the Internet really was linear, which it isn鈥檛,鈥� he says. 鈥淒oing that, you get a wildly low growth projection. The real overall Internet growth rate in terms of hosts is about 100 per cent annually. The user growth rate is similar.鈥�
Quarterman estimates that in October 1994 there were about 55 million people in organisations which had connections to the Internet, but that only a quarter of those people had Internet access. He based this projection on a number of factors including trend tables, and questioning companies directly either over e-mail or the telephone. 鈥淐learly there鈥檚 a huge amount of growth possible without even connecting more organisations, and we know more organisations are connecting all the time. In fact, 200 million Internet users by 2000 is actually a conservative estimate,鈥� he says. Certainly, the growth of commercial companies with Internet links has rocketed. In the US, the number of domains registered under the three standard non-academic suffixes 鈥� .com,.net and.org 鈥� roughly doubled in the three months to the end of December.
Amid all the wrangling over numbers, though, another, larger question occurs: why does it matter? Richard Nuttall, head of business development for Pipex believes having figures is important because 鈥淚t makes a difference when you鈥檙e sitting down in a meeting with someone and you can say that there鈥檚 50 million people out there who they can reach.鈥�
Certainly, the question of how many people are accessible through the Internet is a common question on one of its busier electronic discussion groups, dedicated to Internet marketing. 鈥淚s it possible that many outside the Net don鈥檛 care?鈥� wondered one contributor recently. To which Quarterman replied: 鈥淧robably. But they will have a hard time continuing not to care, considering that the growth of the Internet means it will be where they are pretty soon 鈥� wherever they are.鈥�