Tracking the Global Diffusion of the Internet

Larry Press

Communications of the ACM, pp 11-17, Vol 40, No 11, November, 1997.

Everyone knows the Internet is growing like a weed, but measuring that growth with a degree of precision is difficult. At first, it was easy to follow network diffusion. The Arpanet Completion Report [2] contains maps, topology diagrams, and traffic and performance statistics beginning with a sparse 4-link map drawn in 1969 and running through 1975 when it was turned over to DOD for production. The coming of the Internet made the task more difficult, but the U. S. regional networks and many international networks connected to the National Science Foundation (NSF) backbone, so NSF was able to track and report on traffic and geography [1]. Today, there are roughly 30 national backbones in the U. S. alone, and tracking the global diffusion of the Internet is a daunting, but increasingly important task.

This article surveys some of the organizations tracking that diffusion, and presents some of what they see {footnote 1}. You can consider this entire article a positive review -- the work of each of these organizations is useful and interesting. Some of them focus on the Internet itself while others track its telecommunication and social context, and others measure performance. We will consider each of these categories, beginning with Internet diffusion.

Internet Diffusion

The most venerable student of the global diffusion of the Internet is John Quarterman at Matrix Information and Directory Services (MIDS). In 1986, Quarterman co-authored a major paper [6] global networks (not only the Internet, but the entire Matrix) and followed that with the first book on the topic [7].

MIDS' monthly newsletter, Matrix News ($30/year, $20 for students, $25 and $15 on-line), contains articles on networking developments in various nations and regions, along with excerpts from their more extensive reports. It is mostly written by Quarterman who regularly attends international networking meetings. When not traveling, he is gathering and analyzing Internet data which is published in Matrix Maps Quarterly, MMQ ($400/year, $365 for students, $300 and $275 on-line). The format of each edition of MMQ varies somewhat depending on Quarterman's activity, but it presents both statistics on Internet hosts, international links, and, as the title implies, maps showing links, host counts, and so forth.

The Network Wizards (NW) Domain Name Survey is also among the oldest and most referenced sources of Internet diffusion data. Every six months, NW runs a program that searches the Domain Name System in an attempt to discover every host with an IP address, and they report the number of hosts registered in each top level domain. They also attempt to contact a random sample of the hosts they identify to determine what percent are on line at the time of the test. In January, 1997, they discovered 16,146,360 registered hosts, but 26% of the sample they tested was not reachable.

The NW host counts are widely reported and used by others, often without further analysis. This is unfortunate, because, as NW points out, the number of hosts in a domain is not the same as the number of hosts in a nation. A host registered under a national domain may not actually be in that nation, and, although hosts registered in generic domains like .net are often assumed to be in the U. S., many are not. MIDs begins their analysis with NW data, but refines it. In their most recent analysis, they found that 11 countries have at least 75% of their hosts under domains other than their national top-level domain, and 19 have at least 50% in other domains. Most of these are small nations, but 30.5% of Canadian domains are not registered under .ca, which means NW understates the count by 276,453.

The NW data is quite valuable, since it dates back to August, 1981 (213 hosts), and though the method of collecting the data changed when the DNS came into being, it is a consistent, long- term time series. For highlighting the general trend NW is an excellent source, and if one needs more refined, accurate estimates of national host counts, MIDs is the place to go. (Their world host totals are about 1% apart).

Boardwatch Magazine is also a noteworthy Internet watcher. Begun as a publication for bulletin board system operators, they have made the transition to the Internet and the ISP community. They also publish a bimonthly directory of ISPs ($9.95, 448 pages), and run an ISP convention. The directory was originally restricted to the US and Canada, but has now expanded to include Brazil, and more nations are promised.

As of July/August, 1997, the Directory listed 4,009 ISPs in North America and 192 in Brazil. They disagree with the common wisdom that there will be a great consolidation in the ISP industry, and so far they are right. The current ISP count is up from 3,747 in the May/June directory, and the number of North American backbone providers has risen from 9 to 31 since their first directory in Spring, 1996. The ISPs are also growing. In the Spring, 1996 Directory, they averaged under 2,000 customers and about 12 employees. Today, they average 3,028 customers and 16 employees. Roughly 40% of the Directory is listings of price and service data on the North American ISPs, and the rest is highly informative articles on technology, trends, and, most interestingly, detailed descriptions and topology maps for the major backbone providers in North America Brazil.

Telecommunication Infrastrucutre

The Internet does not exist in a vacuum, and those wishing to understand its diffusion must consider telecommunication infrastructure and social context. Two organizations tracking telecommunication infrastructure are the International Telecommunication Union (ITU) and TeleGeography, Inc.

The flagship publication of the ITU is The World Telecommunication Development Report, (120 Swiss Franks, 217 pages). The 1997 edition is the third in this annual series, and like the others, it is around 70% excellent, well documented analysis articles and 30% statistical tables.

The 1997 articles cover trade in telecommunications equipment and services, international investment, privatization and deregulation, and so forth. There are 21 statistical tables with data on 200 nations. The tables report on telephone infrastructure, wireless communication, data communication, radio and television, equipment trade, global investment, and so forth. For example, they show that high-income, North American nations dominate the installed infrastructure, but growth rates are highest in developing nations (see Table 1). ITU gathers data with a survey of national regulatory agencies, and being self- reported, there may be some bias. Their database also available on-line (updated continuously) and by quarterly subscription on CD-ROM. In addition to the World Report, ITU publishes similarly formatted regional books on Asia Pacific, the Arab States, Africa, and the Least Developed Countries.

TeleGeography, Inc. defines telegeography as "a new branch of geography that maps the pattern of telephone traffic and other electronic communication flows; places created by or perceived solely via telecommunications (e. g., a computer network address); the telecommunications artifacts (radio antennae, terminals, signs) on a site; and the balance of telecommunications power in one country or region vis-a-vis another." Their annual report on international traffic flows, called TeleGeography ($595, $195 for non-profit, government and academic organizations, 174 pages), does a particularly good job on the communication flows and power relationships referred to in their definition. The report is about 40% articles and 60% statistics, and the 1996/7 edition features articles on the leased line markets and technology, the Internet (5 articles), and international facilities and carriers.

Traffic statistics are reported for 73 major telecommunication nations. Between 1985 and 1995, international telecommunication traffic grew from 15.6 to 60.3 billion minutes, and the US, with 22.6 billion minutes is by far the largest communicator (the UK is second with 8.3 billion minutes). International differences are also interesting. For example, most circuits between the U. S. and Canada and Mexico are switched (dial up), and there is little idle capacity, but the majority of circuits to trading nations Singapore and Hong Kong are leased, and there is considerable excess capacity (see Table 2). Leased circuits tend to be used for data and switched circuits for voice traffic.

TeleGeography has several other publications historical statistics and analysis of traffic data, profiles of major international carriers and companies in the computer, software, entertainment, and other related industries, and they also sell data on disk.

Social Context

The World Bank (WB) is an excellent source of economic, demographic, and environmental data. They publish World Development Indicators ($60, 376 pages) annually, and the 1997 edition presents 500 variables from 209 {footnote 2} nations. The book is organized into sections on human capital, environmental sustainability, macroeconomic indicators, private sector development, and global links. The data is presented for the latest year available (1995 in the 1997 edition) and for selected earlier reference years. There is some analysis and graphing, but the data is the "star" of this book.

For those wishing more analysis, WB publishes the World Development Report ($25.95, 265 pages) based on the same data. It is about 70% analysis articles, and 30% statistics. For graphic overview, they publish a smaller Atlas ($15, 48 pages) which presents only a subset of the variables, but contains graphs and world maps for each.

Serious users will want the data on CD-ROM ($275 individual version, $550 networked). The CD-ROM contains 26 years of data and 1,000 preformatted tables. It also has software for retrieving data into spreadsheet-like tables for calculation, graphing, and mapping. While this software is useful for quick, limited views, I soon found myself extracting data from the database for subsequent analysis using spreadsheet and GIS software. This is a good application for component software. Hopefully, in the object-oriented future, the WB will be free to concentrate on what they know best -- data collection and maintenance -- leaving data analysis and presentation to programs specialized in those areas.

Since 1990, the UNDP and Oxford University Press have published an annual report on Human Development ($19.95, 245 pages). The front of the book (60%) is a collection of well-supported essays on a theme (the 1997 theme is Human Development to Eradicate Poverty), and the remainder is statistical tables. Some of the 48 tables in the 1997 edition are for all nations, but many are for either developing or industrialized nations, recognizing different concerns in each group. The flagship index is the Human Development Index (HDI). A nation's HDI is a function of life expectancy, adult literacy, combined secondary and tertiary school enrollment and GDP per capita. Figure 1 shows the relationship between HDI and Internet hosts per capita.

There are also efforts to combine IT and social factors into composite indices of information technology sophistication. The World Times/IDC Information Society Index is the best example I have seen. The ISI is the sum of three sub indices, dealing with social, information and computer infrastructure. These are derived from 19 indicators gathered from sources discussed in this article and some IDC market research. The U. S. tops this index, leading second-place Finland 4,987 to 3,591, and the fastest growing nations between 1996 and 1997 were Japan (18.86%), Malaysia (17.65%), Singapore (16.96%), Korea (15.72%), and Brazil (12.84%).

Network Measurement

Performance, availability, and traffic measurement are important for consumers, ISPs, and network architects. The original ARPAnet contract established UCLA as the network measurement center, and centralized measurement continued through the era of the NSF backbone, but with a profusion of backbones and traffic exchange points in many nations, measurement procedures must be reinvented and decentralized. The Cooperative Association for Internet Data Analysis (CAIDA) has been funded by NSF to deal with this new complexity [4]. CAIDA will develop standard Internet metrics and data formats, and build the tools to gather, analyze, and present the data.

CAIDA will have members, for example, ISPs and equipment manufacturers. Some of the data they gather will be available only to their members, other data will be for sale to customers, and overall aggregate data will be available to the public. This three-tier structure makes CAIDA part industry association, part business, and part government research lab.

It may be argued that government support is unnecessary in a case like this. If ISPs need network measurement, they could fund a trade association for the task. Alternatively, CAIDA's leaders could start a profit-making business to gather and sell information to interested parties. NSF is playing the role of venture capitalist, but they will not retain equity and reap the benefit of a lucrative public stock offering or pass along a tax cut to the citizens.

Will there be public benefit? {footnote 3} The publicly available data will be one concrete benefit. I hope they will regularly post substantial data and analysis geared toward both users and the research community. CAIDA also gives competing ISPs a forum for cooperation in data collection and analysis, which should lead to a better optimized Internet, from which we all benefit. Less tangibly, NSF seed funding gives the public and scientific community a say in CAIDA's management and policy making, and after NSF funding is withdrawn, the organization culture and further sponsored research may keep this influence alive.

While CAIDA's is a coordinated, architected approach, others are already gathering consumer-oriented data. For example, MIDS measures ping (round-trip packet transmission) time between their site in Texas and thousands of locations around the world. They gather data 6 times a day, 7 days a week, and analyze, it producing an "Internet Weather Report" on their Web server. MIDS has been doing this for over three years, and, in spite of greatly increased traffic and predictions of the collapse of the Internet, their average ping time has fallen about 15% per year, though the improvement seems to have leveled off the last few months (Figure 2). While this improvement cannot be generalized to all of the Internet (for example, MIDS' upstream ISP may have upgraded its equipment in the Fall, 1994), it is optimistic.

Boardwatch also has a long-standing interest in Internet measurement for use by ISPs in evaluating upstream carriers and end users in evaluating ISPs. For example, they have publicized MIDS Weather Reports and cataloged traceroute servers which users can use to analyze accessibility to their servers. They recently began a backbone measurement project in collaboration with Keynote Systems.

Keynote offers a website measurement service using programmed agents in 35 North American cities (the number and geographic coverage are growing). For a monthly fee, they will measure download times from a designated Web server to each of their agents, providing data on the performance of the designated server.

Between April 20 and May 20, 1997, Boardwatch used Keynote's service to estimate download time for 50 kbyte documents from the web servers of 29 national backbone providers [7]. Samples were taken every 15 minutes from 27 US cities. The average of all download time estimates from all 29 servers was 9.871 seconds with a standard deviation of 49.061 seconds. The averages for the individual backbones ranged from 1.542 seconds for lightly loaded CompuServe to 26.767 seconds for Bell Canada. While I have reservations about their data analysis and methodology, these can be addressed in the future, and I applaud this first effort, and look forward to more such work. {footnote 4}

Tracking the diffusion of the Internet is a daunting task because it is growing rapidly, is global, and expands organically, at the edges and internally, without central control. Still, business people, policy makers, and capacity planners are better off with approximate data than none at all. Human curiosity and the romance of the whole-earth photo provide less practical reasons for monitoring the global diffusion of the Internet. We are fascinated by the view from space. Over the years, busy humanity has covered the globe with cities linked by railroads, highways, telephone lines, power grids, canals, and so forth, and we are now weaving digital communication links -- the nervous system. I suspect that curiosity and esthetics motivate the people tracking the global diffusion of the Internet as much as profit.


This article owes much to discussion with Sy Goodman and Will Foster.


1. This survey is restricted to reasonably priced studies which are affordable to the research and university community, excluding expensive market research reports.

2. Unfortunately, Chinese political pressure keeps the WB from publishing data for Taiwan. However, they publish data for Hong Kong, and will be allowed to continue after the handover, because it is a "special administrative region."

3. The public has surely benefited from earlier government seeding of networks [5].

4. In particular, the reported standard deviations seem much higher than my experience would indicate. In discussing this with Keynote and Boardwatch, I learned that failed hits were under some circumstances recorded as taking 15-minutes (the time between attempts), skewing the observations. Furthermore, the sizes of downloaded files varied, and the timing estimates were normalized for 50 kbytes. The normalization algorithm may have introduced some error. Finally, some of the variance may be due to server performance, but Keynote is probably correct in their assumption that ISPs would have fast servers directly connected to the company backbone, minimizing this source of variance.


1. Frazer, Karen D., NSFNET: Final Report, Merit Network, Inc., Ann Arbor, MI, 1995,

2. Heart, F., McKenzie, A., McQuillian, J., and Walden, D., ARPANET Completion Report, Bolt, Beranek and Newman, January 4, 1978.

3. Kedzie, Christopher R., Coincident Revolutions, pp 20-29, 53-54/ OnTheInternet, January/February, 1997.

4. Monk, T. and Claffy, K., "Internet Data Acquisition and Analysis: Status and Next Steps," Proceedings of INET '97, Internet Society, Reston, Virginia,

5. Press, Larry, "Seeding Networks: The Federal Role," Communications of the ACM, pp 11-18, Vol 39, No 10, October, 1996.

6. Quarterman, John S. and Hoskins, Josiah C., "Notable Computer Networks," pp 932-971, Communications of the ACM, Vol 29, No 10. October, 1986.

7. Quarterman, John, The Matrix: Computer Conferencing Systems Worldwide, Digital Press, Maynard, MA, 1990

8. Rickard, Jack, "Internet Backbone Measurement Results," Boardwatch, July, 1997, pp 22-53, (reprinted in the July/August, 1997 Boardwatch ISP Directory), and


                   Installed, 1995        94-5 Growth Rates (%)
Income 	       Phone  Mobile   Internet   Phone	 Mobile    Inet
group/         lines  phones      Hosts   Lines	 Phones   Hosts

Lower Income     2.0   0.12        1.35    35.7   135.1   246.0	
Lower-Middle	 9.1   0.33       73.31     8.7   105.1   167.0
Upper-Middle	14.5   1.34      380.13     6.4    66.8   111.9
High	        53.2   8.70    10749.23     3.6    55.6    97.0

Africa	         1.7   0.09       69.14     7.9    60.5    81.4
Americas        29.0   5.17     8359.58     5.4    42.3    91.5
Asia	         5.4   0.62      121.70    14.7   108.3   150.0
Europe	        33.0   3.04     2732.24     3.6    59.5   112.2
Oceana          39.7   9.55    12845.55     4.0    85.7    88.8

World	        12.1   1.56     1661.89     7.0    60.4    97.8

Table 1: Installed base and growth rates for telephone lines, mobile phones and Internet hosts.

Sources: phones, International Telecommunication Union, hosts, Network Wizards.

             Leased    Switched     Idle

Canada         5,543      44,172    1,936
Mexico         1,653      23,416      800
Hong Kong        800         742    1,036
Singapore        521         306      593

World         26,497     126,150  118,343

Table 2. Numbers of leased, switched, and idle 64Kbps circuits to the US.

Source: Telegeography


The Computer Industry Almanac ($53, 788 pages) contains industry statistics, analysis, and directories. Expanded Internet coverage is promised for the next edition. 800-377-6810,

For global background information with an environmental emphasis, see the World Watch Institutes' annual Vital Signs ($12.00, 166 pages, diskette, $89).

Freedom House publishes reports and indices of political rights, civil liberties, and economic freedom. Kedzie [3] has done in- depth analysis of the relationship between their democracy rating and Internet connectivity.

Larry Landweber regularly updates a list showing the showing which nations have IP, Bitnet, uucp, and other connectivity. Current and archived (from 1991) tables and maps are available at

A similar list to Landweber's along with a wealth of information on networking around the world is maintained Olivier MJ Crepin- Leblond at http://

Martin Dodge's Web site has a wealth of information including maps and diagrams of the Internet and many related articles.

For a directory of national and regional Internet Exchange points, see

For NSF backbone statistics through the decommissioning in 1995, see

Worldlink's composite index of information technology combines 5 telecommunication indicators and 3 computer and networking indicators to rank 49 nations.

The World Economic Forum publishes an anual report assessing global competitiveness. http://www/

Regional diffusion and performance data is available at:

   Asia Pacific:
   Latin America:
Organizations reviewed in this article:

   Network Wizards:
   World Bank:
   World Times: