Introduction
What does the Internet look like? Conventionally, engineers have represented it as a cloud, a quick graphic shorthand to mask its complexity. In this paper I consider how cartographic maps and graph visualisations can be used to represent what›s inside the cloud. The paper reviews a range of illustrative projects that have sought to map Internet infrastructure, dividing the discussion into four sections, themed by map purpose: (i) maps for operational Internet management; (ii) maps for Internet marketing; (iii) maps for Internet policy and planning; (iv) maps for academic Internet analysis. Over the last thirty years or so, a huge range of different maps of the Internet have been produced, with diverse forms and function, from simple geographic plans of cable routes to complex real-time 3D visualisations. They have been produced for a number of distinct purposes from planning network deployment, operational management, to prove academic theories, as graduate student projects, for market research, for setting policy and monitoring outcomes, and to try to sell things. And, of course, many have been motivated tomap the Internet for no particular reason other than because it is there. There are many different aspects of the Internet that have been mapped from physical infrastructure, logical layers and protocols, traffic flows, user demographics. The maps cover a range of different scales from individual buildings up to global scale. Many of these maps are beautiful and many more are really rather ugly. A few are actually quite useful, but many more are not very helpful at all. However, all the maps provide a fascinating picture of what the Internet looks like, or rather they provide some insights into what people think the Internet should look once the clouds have cleared.
Power of maps
Conventionally, maps are material artefacts that visually represent a geographical landscape using the cartographic norms of a planar view (i.e. looking straight down from above) and a uniform reduction in scale. They have traditionally been used as static storage devices for spatial data and usually printed on paper, but now they are much more likely to be interactive tools displayed on a computer screen. Today, we live in a map saturated world, surrounded by both conventional geographic maps and many other maplike spatial images and models (e.g. animated satellite images, three-dimensional city models, magnetic resonance imaging scans of the brain). Maps and visualizations have long been used as mode of analysis, providing a uniquely powerful way of making the world more comprehensible. Mapping provides a means by which to classify, represent and communicate information about areas that are too large and too complex to be seen directly. Well designed maps are relatively easy-to-interpret, and constitute concentrated databases of information about the location, shape and size of key features of a landscape and the connections between them. Moreover, the process of spatialisation, where a spatial, map-like structure is applied to data where no inherent or obvious one exists, can provide an interpretable structure to large databases of abstract information.[1] In essence, maps and spatialisations exploit the mind›s ability to more readily see complex relationships in images, providing a clear understanding of a phenomena, reducing search time, and revealingrelationships that may otherwise not been noticed. Here I illustrate the power of a mapping strategy by focusing on its utility in comprehending Internet infrastructure, although mapping and spatialisation can be used to develop an understanding of many different aspects of cyberspace including the structure of the Web and online social interactions.[2]
Maps of the Internet
Internet infrastructure, and its use, is often taken for granted because, unlike roads or railways, it is largely invisible: buried underground, snaking across ocean floors, hidden inside wall conduits, or floating unseen in orbit above us. Indeed, given its invisibility it is easy to assume that it is as ethereal and virtual as the information and communication that it supports. Consequently, there are a number of elements to Internet infrastructure that we presently have little systematic knowledge about, such as the form and function of backbone networks and their subsidiaries, network routing and aggregate traffic conditions, user demographics, marketing penetration and ownership, the physical location of computer servers (hosts) and Internet addresses, connectivity, and bandwidth. The mapping of these elements provide important insights into who owns and controls infrastructure, who has access to the Internet, how the system can be surveyed, and how and from where the Internet is being used. At a basic level, the maps provide a visual inventory and census of where Internet nodes and routes of connection are located, and in specific cases the traffic that flows through networks and their user profiles. Depending on scale, these maps can be used by engineers to install and maintain the physical hardware of the networks, by system operators to manage networks more effectively, and by marketing and business development departments to demonstrate the size and penetration of networked services. In addition, the maps have academic utility by showing significant trends and spatial patterns in the growth of network architecture, service provision, user profiles and traffic flow across spatial scales, so for example, allowing comparison of neighborhoods, cities and countries. As such they reveal the growth of the ‹Network Society› and information economy, but also its uneven and unequal geographic nature by revealingthe distribution of infrastructure and those areas that have poor access to the Internet or are presently excluded altogether. Moreover, they allow an analysis of the changes occurring in these patterns. As recent research highlights, although the Internet has expanded, diversified and diffused greatly, basic infrastructure access and equity issues are still significant; the so called ‹digital divide› issue, which is played out in different ways at different spatial scales, and is fractured along lines of wealth, class, race, gender and so on.[3] The cartographic designs employed are various. Many examples use conventional approaches of shaded or symbol maps on a familiar geographic framework. These are often produced using standard geographic information systems (GIS) packages. However, other significant examples stretch the notion of a ‹map› using more diagrammatic approaches, for example showing the topology of network connections laid out in a non-geographic, abstract coordinate space. Some of the maps are interactive interfaces using the medium of the map to allow users to access and query the data in novel ways. Some of the most potentially powerful and interesting ‹new breed› of infrastructure maps are dynamic in nature, constructed with live data gathered from the Internet every time the map is requested by a user.
Maps for Operational Internet Management
Maps can summarise and present complex, rapidly changing data on the operational state of a network in a single visual image, providing an easy-tointerpret overview of the system and thereby aiding problem diagnosis and solving. For example, in NOCs (network operations centers) of large ISPs just a handful of skilled operations are responsible for keeping a complex and geographically distributed hardware infrastructure running smoothly and maps are essential (see AT&T display).[4] As a New York Times story noted on the huge stress on the U.S. telecommunications systems immediately following the attacks of the 11th September 2001, “By watching computerized maps of the United States, [operators] can tell in an instant whether there are any jams in longdistance traffic.”[5] However, the detailed network monitoring maps and tools used by operators in NOCs are not made public for reasons of security and commercialconfidentiality. Some Internet networks, particularly those serving the research and education communities, do make summary network performance data publicly available using map interfaces. These interfaces are popularly referred to as ‹network weather maps.› These maps represent public-spirited information dissemination tools, providing network customers (usually universities and research labs) with useful information, especially to identify trouble spots, as well as having a marketing function (see next section). Two examples of network weather maps—the Abile network in the US and NORDUnet serving Scandinavia [Figure 3]—are updated frequently (for example the Abilene map is updated every five minutes), allowing users a ‹peak inside› the network cloud. Both maps provide a summary of overall network performance with links colour coded by their traffic flows, but importantly they also provide an interactive, visual interface through which to browse more detailed performance statistics available as tables and statistical charts. These two maps are also illustrative of the two major cartographic archetypes employed to represent computer networks – showing linkages and nodes either as a logical schematic diagram or on a geographic base with a familiar template of cities and administrative boundaries. These maps can often be highly generalised, with for example the network architecture shown as straight lines, although they are topologically correct (as with conventional subway maps). One useful method, available to average Internet users, for monitoring network performance is traceroute, which allow the active ‹probing› of real-time data routing and speed. Traceroutes are simple utility programs which report the route data packets travel through the Internet to reach a given destination, and the length of time taken to travel between all the nodes along the route. Designed primarily for network engineers to ‹debug› routing problems, they are also useful tools for researchers to scan the inside of the Internet cloud. They reveal the hidden complexity of data flows, showing how many nodes are involved (often more than twenty), the seamless crossing of oceans and national borders and the multiple transfers through networks owned and operated by competing companies. They can also detail how geographically illogical some data routing is,following the cheapest paths rather than the shortest.[6] —see a typical text-based output of the basic traceroute utility. Each line in the output of traceroute represents a single ‹hop› the data takes through the Internet. In this case the data route took 30 hops to reach its destination. Each hop is generally a separate physical node comprising of dedicated switch or router hardware. The approximate locations of this routing hardware can also be plotted on a map to give a geographic traceoute (see the VisualRoute utility).
Maps for Internet Marketing
A large number of infrastructure maps of the different Internet networks have been produced primarily for the purposes of marketing. Indeed, a cursory examination of most any ISP websites will reveal ‹high-gloss› marketing maps. This is, perhaps, not surprising as maps have long been created in the service of marketing and promotion.[7] Geographic maps can be seen in some senses as the natural visual representation of transportation and communications networks, able to effectively show potential customers how a particular network could expedite their travel needs. As a consequence, there is a long (dis)honourable tradition of promotional maps being used to highlight the advantages of the latest transportation network such as canals, oceanic shipping lines, railroads, highways and of course airlines.[8] Given that the provision of Internet network services is a highly competitive business, dominated by large corporations many of whom operate globally, effective marketing is a vitally important activity. Here, maps are employed to provide a selective and positive view of a network, emphasizing its extent (e.g. demonstrating the geographic reach of the network, emphasizing all the distant places that are linked together) and capabilities (e.g. illustrating the tremendous capacity of the ‹pipes› of the network to cope with huge users demands) in order to attract and compete for custom. In many respects Internet network provision is such an intangible commodity that the map is powerful in making it seem more ‹real.› The maps generally show a generalized and simplified view of the network, usually in a bright, colorful and visuallyeffecting manner. Most often the maps are drawn on a template of real world geography and have many design commonalties with the airline route maps in the back of in-flight magazines. While these maps do provide a selective picture, a reflection of what the company wants to emphasize, they also allow academic researchers and market researchers to chart the range and make-up of each companies network, to document different kinds of provision at a range of scales, and importantly to note how this has changed over time. This can be illustrated in reference to an analysis of UUNet's infrastructure (formerly part of Worldcom). Infamously, it was claimed in the late 1990s that the UUNET network was growing at rate of 1000 per cent per year. A longitudinal study of their marketing maps at a variety of scales allowed researchers to see the company’s strategy for delivering infrastructure services and to project the likely consequences this strategy on issues such as the digital divide, urban-regional restructuring, local and regional economic development, and so on). In fact, the 1000 % growth figure was apocryphal[9] and was an element in the «dotcom» hype of the late 1990s that led to significant over investment in fibreoptic infrastructure.
Maps for Strategic Planning And Policy
Maps have been key strategic devices used in planning and implementing urban and regional development, plotting military strategy and the conquest of new lands, and legally contesting land ownership and use. Unsurprisingly then they are also being used in the short and long term strategic planning of Internet development by commercial enterprises, governmental, quasi-governmental, and other interested bodies (e.g. the Internet Society). In order to structure the analysis I have divided the discussion into two related themes. The first concerns the planning and development of infrastructure, the second, regional development, the attraction of inward investment, and the monitoring and addressing of inequalities. At one level, maps have been used in the planning, development and expansion of network infrastructure at a variety of scales from individual buildings to global networks. Planning the optimum topology for a communications network to efficiently interconnectgeographically dispersed locations is an exacting task. Maps help visualize complex network topologies and how new configurations will look and operate—see the ‹back of the envelope› hand drawn sketch map from the early planning of ARPANET[10] and the fibreoptic cable routing in downtown Philadelphia. By mapping companies in relation to cable-routing the city can adequately provide network connections and plan extensions that will hopefully attract in new customer. At a larger-scale, countries are crisscrossed by many interconnected networks. An important function for ISPs is to easily and efficiently interconnect and exchange local traffic at neutral peering points as shown by examples of national-level maps tracking the Internet infrastructure in the Republic of Korea. The maps are valuable policy and research resource creating a census of the growing complexity of the links between ISPs and their capacity. At a second level, maps have been employed in the strategic planning and implementation of regional development and in monitoring and addressing inequalities, the so-called digital divide, between places. Again, the data relates to several scales from intra-urban to global. As widely documented, cities are increasingly becoming competitive enterprises, vying to attract investment of the high-tech sector. Maps are a potentially important tool for illustrating highcapacity internet infrastructure to potential inward investors and encouraging economic development. Examples include the «Bandwith Bay Fiber Network Mapping» by the City of San Diego and the «Georgia High-Speed Telecommunications Atlas» as well as document the so called ‹knowledge economy.›[11] (See the variations in the home Internet access across London) Similar quantitative assessment and mapping of the geographical patterns of competitive in the knowledge economy has also be undertaken at a national scale. Mark Krymalowski analysed data at the country level, plotting the geographical distribution of .DE domain registrations in Germany. In other words, the information economy is likely to grow most quickly around existing IT hubs, rather than invest in new, potentially cheaper, locations. These kind of maps when put together in a timeline, form a powerful means for tracking development and for predictingfuture change. One project that illustrates this is that by Larry Landweber, and several organisations have taken his lead to produce longitudinal maps at different scales (e.g. TeleGeography). During the first half of the 1990s the Internet spread across the globe so that by the end of the decade virtually all nations were connected (although the number and capacity of connections still varies greatly). This global diffusion of the Internet was tracked by Landweber and charted in a series of maps providing a useful baseline census for policy of the spread of international network connectivity.[12]
Maps for Academic Internet Analysis
The seemingly magical ability to surf around a virtual globe of information, moving from website to website at a single click, belies the scale and sophistication of the sociotechnical assemblage of protocols, hardware, capital and labour that makes this possible. It has been widely argued by academics that the ICTs are transformative technologies that are having significant impacts on social, economic and political life, engendering widespread changes e.g. in relation to urban-regional restructuring as illustrated above.[13] The process of mapping has been used to comprehend two other sorts of projects aimed at furthering our understanding of these changes in relation to infrastructure: the digital divide; measuring the Net. As noted above, maps reveal visually the nature and extent of the ‹digital divide› in society. [14] Matthew Zook has analysed the spatiality of the Internet content production industries in the US through the detailed mapping of the geographic location of domain name registrations at different scales. Just as postal addresses in the geographic space identify a unique location, domain names perform the same function for the Internet, allowing users to visit the site. Importantly, the geographic location of the owner of these domains can be determined from registration databases, which have a billing postal address, containing zipcodes that can easily be mapped to street-level locations using offthe- shelf GIS software and map data. This mapping led Zook to conclude that the ‹Internet industry exhibits a remarkable degree ofclustering despite its reported spacelessness.›16 This approach provides a valuable quantitative measurement for policy analysis on Internet economic activity and revealing where is connected and just as importantly where is not. Despite the virtualised rhetoric, this assemblage remains embedded in real places and maps can help to reveal the intersections between cyberspace and geographic space. In academic Internet research, an understanding of that geography is important, as knowledge of the physical location of virtual phenomena can tell you interesting things (such as which territorial jurisdiction it is in) and can also enable the linkage to a large array of existing secondary data (for example socio-economic characteristics from censuses). The ‹where› and ‹how› of the physical embeddedness of data networks and information flows is also important because of their uneven global distribution and the consequent socio-spatial implications in terms of access and inequalities, as starkly revealed in. This is a global scale mapping of network infrastructure that contrasts the density of core Internet routers with the distribution of population. The maps are density surfaces, where the land is colourcoded so that higher densities are darker. In design terms they are really quite conventional cartographic maps, using a geographic framework of continental outlines to show univariate data. This type of world map is familiar to most people and can be easily produced using GIS software, and succinctly summarises a large volume of data in an intuitive manner. The final way that maps have been used by academics and commercial research teams is a means by which to display measurements that quantify the extent and use of Internet infrastructure so as to gain a better understanding of its distribution, diffusion and utilisation. [15] In an ‹arc map› of Internet traffic flows between fifty nations, from February 1993, the colour, thickness and height of the arcs are used to encode the traffic statistics for particular inter-country routes.[16] In the SeeNet3D application in which the image was generated, the user had considerable interactive control able, for example, to vary the arc height, scaling and translucency. The map could also be rotated and scaled, so that the user can view it from any angle. The map shows that there was significant traffic, in the early 1990s, betweenthree areas of the world, North America to Europe, Europe and Australiasia, and Australiasia and North America, with most traffic crossing the Atlantic. The map does not show all traffic, however, because it is limited to just fifty countries. As such, it portrays a selected image, one that is dominated by developed countries that were the principle nations connected to the Internet in 1993. The final example is the Internet Mapping Project being undertaken by Hal Burch and Bill Cheswick at Lumeta Corporation.[17] Their project maps the topology of thousands of interconnected Internet networks to provide perhaps the best currently available large-scale overview of the core of the Internet in a single snapshot. They map the Internet in an abstract space (i.e. using a process of spatialisation), thus disregarding the actual location of nodes in physical space. Data is gathered by using the Internet to measure itself on a daily basis, surveying the routes to a large number of end-points (usually Web servers) from their base in New Jersey, USA. The resulting spatialisation maps how hundreds of networks connect together to form the core of the Internet. In the example shown, links have been colour-coded according to the ISP, seeking to highlight who ‹owns› the largest sections of Internet topology. This project is ongoing and the data is archived and available to other researchers to utilise. Over time, it is hoped that the data will be useful for monitoring growth and changes in the structure of the Internet. The experience gained in mapping the Internet is also being applied commercially, using network scanning and visualization techniques to chart the structure of corporate intranets to identify security weaknesses and unauthorized nodes.
Conclusions
I have argued in this paper that mapping can be used to look «inside the Internet cloud,» providing a significant analytical tool for managing Internet infrastructure, developing and implementing policy, and understanding the information economy. Maps can be used to reveal the range, extent and density of Internet infrastructure in relation to realworld geography at a variety of scales. I finish on a note of caution, however. While mapping is a useful strategy, with many of the maps visually striking and persuasive,they need to be created, used and interpreted with care for several reasons. Firstly, map-making is now much easier to do, but it is not necessarily a quick fix to understand the Internet. Like any 19 chosen mode of analysis, the potential of mapping has external practical constraints, including data availability and data quality and the level of user knowledge. There are also issues to consider relating to the ethics and responsibility of researchers producing maps of the Internet. The processes of data selection, generalisation and classification and the numerous map design decisions mean that one can never remove the subjective element in map-making. As Monmonier notes: «… any single map is but one of many cartographic views of a variable or a set of data. Because the statistical map is a rhetorical device as well as an analytical tool, ethics require that a single map not impose a deceptively erroneous or carelessly incomplete cartographic view of the data. Scholars must look carefully at their data, experiment with different representations, weigh both the requirements of the analysis and the likely perceptions of the reader, and consider presenting complementary views with multiple maps.»[18] Further, some of these new mappings of the inside of the Internet cloud can also been seen as a new kind of surveillance, revealing connection and interactions that were previously hidden in unused log files and incomprehensible databases. The act of mapping itself may constitute an invasion of privacy. If the appeal of the Internet is its aspatial anonymity, then users may object to it being placed under wider scrutiny, even if individuals are unidentifiable. Here, public analysis may well represent an infringement of personal rights. In some senses, these maps may work to shift the spaces they map from what their users consider semi-private spaces to public spaces, and thus the maps may actually change the nature of Internet itself. Therefore, it is important to consider the ways, and the extent to which, maps of the Internet cloud are ‹responsible artefacts,› that do not destroy what they seek to represent or enhance. Lastly, it should be recognised that mapping is also cultural process of creating, rather merely revealing knowledge. All the sophisticated, interactive maps of network infrastructures have politics just the same as any other form of cartographic text, and we must bealert to their ideological messages.[19] Maps of the Internet cloud can prove to be very valuable, but at the same time they can never be value-free. There is no one single map or technique that can capture all the complexities of the Internet cloud. Instead, there are a multiplicity of different Internet maps that focus on different components of the infrastructure. Perhaps, even, our knowledge is diminishing as the scale and complexity of infrastructure grows and information about it becomes less open to scrutiny. So as the Internet cloud grows evermore expansive and dense, it is becoming even harder to see the details within.