Download Geographic Information Science and Distributed Computing: Implications for GIS and more Summaries Mobile Computing in PDF only on Docsity!
D ISTRIBUTED AND MOBILE C OMPUTING
Michael F. Goodchild^1 , Douglas M. Johnston 2 , David J. Maguire^3 , and Valerian T. Noronha 4
Abstract
Technology is moving rapidly to the point where computing will be available everywhere, will be fully mobile, and will provide access to widely distributed resources. This trend to itinerant, distributed, and ubiquitous computing is the latest in a long series of major architectural changes, with associated implications for where computing is performed. Production of geographic data is similarly moving to a new set of arrangements focused on local agencies and individuals. These changes are placed within an economic framework, as a basis for development of a new set of theoretical principles regarding the location of computing, and its implications for geographic information science. Digital libraries are a further instance of these trends, with associated implications for the spatial organization of libraries. The chapter ends by identifying a series of research issues, and the benefits that can be anticipated if these issues are solved.
1. I NTRODUCTION
Over the past few years a large number of advances in computing and communications technology have made it possible for computing to occur virtually anywhere. Battery- powered laptops were one of the first of these, beginning in the mid 1980s, and further advances in miniaturization and battery technology have reduced the size of a full- powered but portable personal computer dramatically—in essence, it is the keyboard, the battery, and the screen that now limit further miniaturization and weight reduction in the laptop. More recently, the evolution of palmtop computers and other portable devices (Portable Digital Assistants or PDAs), as well as enhanced telecommunication devices have further stimulated the trend to mobile computing. The evolution of new operating systems (MS Windows CE and JavaSoft Java OS) and software components (MS COM and JavaSoft Java Beans are the main standards) have also had major impacts. The Internet has vastly improved inter-computer connectivity, making it possible for people to share data, visualizations, and methods while separated by great distances. Wireless communication technologies, both ground- and satellite-based, now make it possible to connect from places that have no conventional connectivity, in the form of copper or fiber. In short, we are rapidly approaching a time when computing will be:
- itinerant , maintaining full connectivity and computational power while moving with a
(^1) National Center for Geographic Information and Analysis, and Department of Geography, University of
California, Santa Barbara, CA 93106-4060, USA. Phone +1 805 893 8049, FAX +1 805 893 3146, Email good@ncgia.ucsb.edu (^2) Departments of Landscape Architecture and Geography, University of Illinois, Urbana, IL 61801, USA. (^3) Environmental Systems Research Institute, 380 New York St, Redlands, CA 92373, USA. (^4) Digital Geographics Research Corporation, 31-3600 Colonial Dr., Mississauga, Ontario, Canada L5L
5P5.
person, in a vehicle, or on an aircraft or ship;
- distributed , integrating functions that are performed at different places in a way that is transparent to the user; and
- ubiquitous , providing the same functionality independent of the user's location.
Of course we are still some way from achieving this ideal, and it is still easier and cheaper to compute in the conventional manner at a desktop using only the data, software, and hardware present in the workstation. Wireless communication is still more expensive, and more constrained by bandwidth, at least in those areas of the developed world that are well-served by copper and fiber. But the pace of change is rapid, and in some countries we are already approaching the point where the majority of telephone communication will be wireless.
New devices and methods of communication suggest the need for a fundamental rethinking of the principles of geographic information systems design. Computers are likely to be used in new kinds of places that are very different from the conventional office; to communicate information using new modes of interaction, such as the pen; and to require much better use of restricted displays and computing power. On the other hand, devices are already available that offer the power of a desktop machine in a wearable package. All of these developments challenge the research community to investigate entirely new applications, and to address technical problems.
This discussion is motivated largely by technological change, for three major reasons. First, technological change induces change in society, its practices and arrangements, and in the conduct of science, and it is important to anticipate such changes. By moving quickly in response to anticipated change, society and science will be better able to take advantage of the benefits of new technology, following the pattern that has typified the history of geographic information systems over the past three decades (for histories of GIS and the role of technology in their development see Coppock and Rhind, 1991; Foresman, 1998). Second, it is to be expected that generic technologies such as those of itinerant, distributed, and ubiquitous computing (IDU for short) will require specialized adaptation to exploit their potential fully in different domains. The geographic domain is sufficiently distinct and complex that substantial research and development activity will be needed, and substantial efforts will have to be made to introduce these new technologies into the domain's many application areas. Finally, research will be needed to explore the impacts these new technologies will have, and the issues they will raise, in areas such as personal privacy, community-based decision-making, and the accumulation of power.
IDU computing is by its nature generic, so it is not immediately clear that it merits special attention by the geographic information science community, or that substantial research problems exist that if solved will have value to the GIS application domain. But two arguments seem especially compelling in this regard. First, all three characteristics of IDU computing—the ability to be itinerant, distributed, and ubiquitous—are inherently geographic. Mobility specifically means with respect to spatial location; there are clear advantages to being able to integrate computing functions across many locations; and being able to compute anywhere in space is clearly an advantage. It is implicit in this chapter, therefore, that IDU refers to capabilities in space, rather than to capabilities in
2. THE NATURE OF COMPUTING
2.1 What is computing?
Computers began as calculating machines, designed to process very large numbers of arithmetic operations at speeds far beyond the capabilities of humans using tables of logarithms, slide rules, or mechanical calculators (Wilkes, 1957). Massive calculations had become necessary in many human activities, including surveying, nuclear weapons research, physical chemistry, and statistics, and the computers that began to appear on university campuses and in government laboratories in the late 1940s and 1950s permitted very rapid advances in many areas of science and engineering. Cryptography provided another well-funded and highly motivated application, in which computers were used not for arithmetic calculations but for very large numbers of simple operations on codes. In essence, the development of computing was the result of a convergence of interests between the military and intelligence communities that flourished in WWII and the early days of the Cold War, and the more general needs of science. Languages like FORTRAN reflected these priorities in their emphasis on calculation, and the representation of mathematical formulae in computer languages.
Until the 1980s the community of computer users was small, and the vast majority of the population had little need to interact directly with computing machinery. Even on campuses, the community of users formed a small elite, with its own largely inaccessible language, unusual hours of work, and seemingly endless demands for funds to purchase computing time. But the advent of the personal computer changed this dramatically. Academics with no interest in calculation suddenly found the need for a computer; and computers entered the classroom, in areas far removed from science and engineering. Computers entered the family household, and became a significant source of entertainment. Today, of course, numerical processing occupies very little of the capacity of society's computers, especially if weighted by number of users, and very few users are concerned with the code transformations of cryptography. Instead, computing is about accessing information, presenting it in visual form for easy comprehension, searching through databases, and sending information to others. Early computing was dominated by processing and transformation in the service of the user; today it is dominated by communication , either between users, or between the user and the machine as information storehouse.
Today, one computes if one uses a computing system to:
- perform a series of arithmetic operations (e.g., run a population forecasting model);
- display or print information (e.g., print a map or driving directions);
- transmit information to another computer (e.g., share a database with another user);
- search a network of computers for information (e.g., browse the World Wide Web);
- request information from a local or remote database (e.g., download an image from an archive);
- transform information using a set of rules (e.g., convert data from one map projection
to another);
and many other possibilities. All of these in some way involve communication of information, possibly combined with transformation, between computers and users. Users may communicate remotely with computers, and computers may communicate remotely with each other.
2.2 The location of computing
In the early days of computing there were no communication networks, and there were very strict limitations on the possible distance between the input and output peripherals, typically card readers and printers, and the central processing unit. While a user could travel away from the location of the input and output peripherals, there was a distinct cost to doing so: input and output had to be communicated physically, by shipping cards, tapes, or printed output, and thus incurred a substantial time penalty. Nevertheless this was often considered worthwhile, since computing capacity was so expensive relative to the costs associated with delay. Thus a user might willingly suffer a one week delay in order to obtain the processing services of a computing system at a remote campus.
Today, of course, such delays are no longer normal or acceptable. Computing has become so ubiquitous and cheap that delays are rarely experienced, and one expects to be able to connect to significant computing resources from almost anywhere. The locational pattern of computing has changed substantially in thirty years.
Nevertheless, every bit of information and every operation in a central processing unit has a well-defined location, at the scales of geographic experience (Heisenberg's uncertainty principle applies only at scales well below these). It is clear where the user is located, where bits are located on hard drives, where communications networks run, and where transformations of data occur as a result of the operation of software. From the perspective of communication, the important locations include:
- the location of the user, where information finally resides as a result of computing;
- the location of the user's workstation, which may travel with the user or may be fixed at some defined location such as an office desk;
- the locations of the network used to transmit information to and from the user's workstation;
- the locations where information is transformed or processed into the form requested by the user;
- the locations where the necessary data are stored;
- the locations where the data are input to the network or storage locations;
- the locations where data are interpreted, processed, compiled, or otherwise prepared for storage;
- the locations where the data are defined or measured;
- the locations that are represented, in the special case of geographic data.
The last two bullets are of course particularly relevant to geographic data, since such
- interest : since geographic data are most likely to be of interest to users within the spatial extent or footprint of the data, they are most likely to be served from a location within that spatial extent;
- legacy : since data servers often replace traditional services, such as map libraries or stores, they often inherit the locations of those services, and the institutions that sponsored them.
Nevertheless, with the trend to out-sourcing and facility management contracts it is sometimes the case that computing occurs at a third-party location unaffected by any othese factors.
Each of the other tasks listed above also has its associated locational criteria. Compilation of geographic data often occurs in public sector agencies, such as the U.S. Geological Survey, at its national headquarters or at regional facilities. Increasingly, it is possible and cost-effective to compile geographic data locally, in county offices of the Department of Agriculture, or in City Halls, or even in the farm kitchen. Ultimately, such locations are constrained by the locations of the human intelligence that is an indispensible part of the compilation process. But with wireless communication and portable hardware, that intelligence may be best located in the field, where phenomena are best observed and where observations are most comprehensive and uninhibited.
In a world of IDU computing, therefore, the locational patterns of computing are likely to adapt to those of the human institutions and intelligence that define the need for computing and use its products. This is in sharp contrast to the historic pattern, when the scarcity and cost of computing were the defining elements. Although computing in an IDU world can occur anywhere, its various stages and tasks must occur somewhere, and there are both economic and less tangible criteria to provide the basis for locational choice. Exactly how, and with what long-term consequences, is a research issue that geographic information science should solve, so as to anticipate the long-term effects of IDU computing on the distribution of human activities. IDU computing may alter the locational patterns that evolved prior to the use of computers to communicate information; or further alter patterns that adapted to earlier and less flexible forms of computing, such as the mainframe.
3. THE LOCATIONAL HISTORY OF COMPUTING
IDU is the latest of a series of forms of computing technology that have followed each other in rapid succession since the early days of computing in the 1940s. In this section three phases of development are identified, each with a distinct set of locational imperatives.
3.1 Phase I: the single-user mainframe
From the 1940s through the mid 1960s computing technology was limited to mainframes, each costing upwards of $1 million, and financed by heavy charges on users based on the number of cycles consumed. Each user would define a number of instructions, to be executed in a batch during an interval while the user had been granted control of the
machine.
High-speed communication was expensive and limited to very short distances. In essence, then, the locations of computers were determined by the distributions of their users in patterns that are readily understood within the theoretical framework of central facilities location theory. In classical central place theory (Berry, 1967; Christaller, 1966; Lösch, 1954)) and its more recent extensions, a central facility exists to serve a dispersed population if the demand for the service within its service area is sufficient to support the operation of the service. The minimum level of demand needed for operation is termed the threshold , measured in terms of sales for commercial services or size of population served for public services. The distance consumers are willing to travel to obtain the service or good is termed its range.
In principle, mainframes were spread over the geographic landscape in response to the distribution of demand. In practice, this demand was clustered in a few locations, such as university campuses. A few users not located in such clusters were willing to travel to obtain computing service, or to wait for instructions and results to be sent and returned by mail, because no alternative was available. The pattern of mainframes that emerged was thus very simple: one mainframe was located wherever a cluster was sufficiently large or influential to be able to afford one. In time as demand grew and the costs of mainframes fell it became possible to locate multiple mainframes where clusters were sufficiently large. In summary, the characteristics of Phase I were:
- very high fixed costs of computers;
- a highly clustered pattern of users;
- costs of travel for those users not located in large clusters that were low in relation to the costs of computing.
These conditions became increasingly invalid beginning in the mid 1960s, and today it is difficult to identify any legacy of this period, with the possible exception of certain large scientific data centers which still occupy locations that were determined initially by the presence of mainframes, and which still benefit from substantial economies of scale through co-location of staff and servers.
3.2 Phase II: the timesharing era
By the mid 1960s, developments in operating systems had made it possible for computers to handle many users simultaneously, through the process known as time-sharing. Although very large numbers of instructions could be executed per second, only a fraction of these would be the instructions issued by one user. As a result, it became possible for users to issue instructions and receive results interactively over an extended period of time, without being constrained to batch operation. This mode of operation required only relatively slow communication speeds between users and computers, speeds that could be supported by existing teletype technology over standard telephone lines. Terminals , consisting initially of simple teletype machines and evolving into combinations of keyboards and cathode ray tube displays, provided for local input and output. More sophisticated displays appeared in the 1970s that could display simple graphics. Only batch and advanced graphics applications required high-speed
3.4 Phase IV: the networked era
Although the idea of connecting computers had its origins in the 1950s, and although many of the technical problems had been solved by the 1960s, high-speed communication networks capable of carrying bits at rates far above those of conventional telephone networks finally became widely available at reasonable costs only in the 1980s, with the widespread installation of fiber, microwave, and satellite links. Fiber networks were installed on university campuses, through industrial sites, and between major cities, although connections to the individual household are still largely restricted to telephone networks operating at a few thousand characters per second. Internet technology permitted bits to flow over these hybrid networks in ways that were essentially transparent to the user.
Computer applications evolved quickly to take advantage of the development of networking. Client-server architectures emerged to divide computing tasks between simple client systems owned by users, and more powerful servers owned by a range of providers. For many applications, software and data could reside with the server, while instructions were provided by the client and results returned. The World Wide Web represents the current state of evolution of client-server technology, with powerful servers providing services that range from sales of airline tickets and information about movies to geocoding and mapping.
Today's communication networks present a very complex geography (Hayes, 1997). A computer that is located at a node on the network backbone may be able to communicate at speeds of billions of characters per second, while another location may require telephone connection to a high-speed node, restricting communication to a few thousand characters per second. Other locations may lack fixed telephone connection, so communication will have to be carried over wireless telephone links. The most remote locations will be outside the coverage of wireless telephone systems, and will have to use relatively expensive and unreliable satellite communication. Economies of scale are highest for the backbone, resulting in very low costs per user or per bit; while the costs per user or bit of the so-called last mile or most peripheral connection may be far higher. The location theory of networks (e.g., Current, 1981) provides a comprehensive framework for optimization of communication network design.
In summary, four phases of evolution of computing have completed a transition from a location pattern based on provision of service from point-like central facilities of high fixed cost, to a pattern of almost ubiquitous, low-cost facilities located with respect to a fixed communications network. In Phase I, computers established themselves wherever a sufficient number of users existed; in Phase IV it is connectivity, rather than the existence of users, that provides the most important economic determinant of location, along with a large number of less tangible factors. Over the forty-year interval the costs of computing have fallen steadily; in the context of GIS, the cost of hardware and software to support a single user has fallen from around $100,000 to $100 in today's dollars.
The next section discusses the nature of communication technologies in more detail, and also introduces new technologies of computing that extend its significance in IDU applications.
4. IDU TECHNOLOGIES
4.1 Communications
As noted earlier, the first widely available form of communication with computers was provided by a simple adoption of existing teletype technology, which allowed characters to be sent over standard telephone lines using ASCII code, at rates of a few hundred bits per second (essentially determined by the rates of typing and printing achievable in mechanical teletypes). The coded signals of the teletype were made compatible with the analog technology of telephone networks by use of modems (modulator/demodulator), which converted streams of bits into acoustic signals. Today, rates as high as several thousand characters per second have become routine through incremental improvements in telephone systems.
Local area networks (LANs) are communication systems that connect computers over limited geographic domains, such as a single office complex or a campus, using combinations of copper wire and fiber, and achieving rates of millions of bits per second. They are largely transparent to users, to whom the network appears as an extension of the desktop workstation, with additional storage devices and processors.
Over the past ten years the Internet has become the most widely known instance of a wide area network (WAN), with services that approach those of the LAN in connectivity and transparency. Indeed it has become common to compare the Internet and the computers connected to it to a single, vast computer. Internet services such as the WWW provide additional layers of functionality.
At the periphery, connection is provided by a series of technologies that are much less reliable, and less and less transparent to the user. When computers move out of the office and into the vehicle or field, data communication can take place only over wireless media. This currently presents implementation hurdles, increased cost, and constraints on data communication rates. But the field of wireless communication is developing rapidly, both in the technologies available, and in the number of subscribers. In developing countries where the telephone cable infrastructure is not extensively developed, wireless voice telephony is attractive, and increasingly cost-competitive with traditional wire line. It is reasonable to expect that within a decade, wireless technologies will be far more advanced and more readily available, facilitating an explosion of IDU computing.
4.1.1 Technology Wireless communication relies on radio waves, which are part of the spectrum of electromagnetic radiation (microwaves, visible light and x-rays are examples of waves in other ranges of the spectrum). The frequencies currently employed for data communication are about 300 kHz to 6 GHz, corresponding to wavelengths of 5cm– 1000m. Weak electrical signals—music broadcasts, taxi dispatcher instructions, GPS transmissions from satellites, or cell phone conversations—are loaded onto stronger radio waves by a process called modulation. The wave is transmitted at a certain frequency; a receiver tunes in to this frequency, demodulates the wave and retrieves the electrical signal. The signal fades while traveling through the air, weakened by interference with other radio waves, and confused by bouncing off physical obstacles. Intermediate
voice and data communication. An area is organized into honeycomb-like cells, and a base station in each cell operates a transceiver (transmitter and receiver) with an operating radius of 10–50 km. Micro-cells and pico-cells can be set up in specific zones such as tunnels and buildings. Mobile phones communicate with the transceiver, which in turn is connected to the wired network. As a mobile unit nears the cell boundary, the base station senses a reduced signal strength, and hands off control to the neighboring cell. Within a fraction of a second, the frequency of communication changes, and the call resumes, the switch being transparent to the user.
The first wireless telephones were offered in the 1940s, but the concepts of cells and frequency re-use were developed later, and it was only in the late 1970s that automatic switching was sufficiently developed, and licensing authorities permitted widespread public use. North American service began in 1983 with the ~800 MHz Advanced Mobile Phone System (AMPS), which remains the most popular service. It is primarily an analog system, but recently there have been digital outgrowths.
Analog Cellular. To transmit digital data over an analog network, the data are modulated into audio signals (the “chirps” heard when a modem or fax machine establishes a connection); the audio is then transmitted in exactly the same way as a voice. Transmission rates are low, in the 2400–9600 bits per second (bps) range, and analog is subject to fading and other problems described above.
An alternative was introduced in 1993: Cellular Digital Packet Data (CDPD) or Wireless IP. CDPD is carried mostly over the analog cellular network. Data are organized into Internet Protocol (IP) packets , which are transmitted in short bursts over analog lines during lulls in voice traffic. Transmissions hop between channels in search of vacant slots. CDPD is currently one of the most effective forms of wireless data exchange, particularly for intermittent transmission, up to 1 kb at a time (circuit switching is more appropriate for continuous data communication). CDPD operates at 19.2 kbps; actual throughput rates are 8–11 kbps. Encryption, combined with channel hopping, make CDPD extremely secure. Service is billed by the number of packets transmitted, rather than air time, making it particularly appropriate for multiple mobile units (e.g. vehicle fleets). CDPD also operates on digital networks, on dedicated frequencies.
Digital Cellular. Digital service is new in North America, largely because the continent was relatively well served by analog AMPS in the 1980s; by contrast, in Europe, multiple analog protocols were implemented simultaneously, and a lack of standards inhibited interoperation between countries. Europe therefore took an early lead in the switch to digital technology. The Groupe Speciale Mobile (GSM) established a standard that has now been adopted in many countries. Unfortunately GSM is only marginally compatible with AMPS. Hence North America went a different route with digital cellular in the 1990s, employing voice digitization and multiplexing to create Digital-AMPS (D- AMPS), which was backward-compatible with AMPS. There is also some GSM in America, under the label PCS-1900. GSM is currently the only digital cellular technology in America that supports data transmission (D-AMPS does not), albeit at a relatively slow 9.6 kbps. In this context it is worth noting that the term PCS is used with liberty, and that some services sold under the title PCS are based on D-AMPS technology. Figure 1 summarizes the cellular options in North America today.
Audio
Data
AMPS
CDPD
GSM (PCS-1900)
Analog
Digital
Hybrid (principally digital)
0100111001
D-AMPS
Figure 1. Cellular options in North America. AMPS is the most widely available.
Spread spectrum. The principle of spread spectrum (SS) is to spread a signal very rapidly over a wide range of frequencies (similar to frequency hopping in CDPD), according to a specific pattern. If a receiver knows this pattern, it can de-spread the signal and recover the data. Current technology allows frequency hopping at a rate of about 40 per second. The method is resistant to jamming and interference, and is inherently secure because the receiving party must know the frequency hopping schedule. For this reason, although SS was originally developed about 50 years ago, it was largely restricted to the military until about 1985. It is now growing rapidly because of the high data rates it enables, and because it allows multiple users to share the same frequency space. Wireless wide and local area networks are being developed based on spread spectrum, and it is being proposed as a generic means for users to access Information Service Providers (ISPs). The drawback is cost, which is currently high. Three radio bands are now reserved for spread spectrum, at approximately 900 MHz, 2400 MHz, and 5.8 GHz, and licensing requirements are minimal.
FM sub-carrier technology is the basis of the Radio Data System (RDS) in Europe, or Radio Data Broadcast System (RDBS) in the U.S. Just as CDPD makes use of unused voice cellular bandwidth, RDS exploits an unused portion of commercial FM broadcast waves. Unlike cellular and spread spectrum, this is a one-way broadcast technology, with data targeted to subscribers to an information service, rather than to the individual user. RDS is used for broadcasting digital traffic reports from Traffic Management Centers (TMCs), hence the acronym RDS-TMC. A decoder in the vehicle can be programmed to filter out traffic messages that do not pertain to the current route, and to accept text data transmissions only in a given language. RDS also has a feature to override an audio station, and to wake up a radio from sleep mode. It is therefore well suited to disaster warning. Another popular RDS application is real-time differential GPS correction, where 100 MHz RDS competes against 300 kHz radio beacons.
RDS is not yet widely established in the U.S. There are message standards issues to be resolved, and location referencing in particular is an area in which standards are still under development (e.g., Noronha et al. , 1999). Currently there are two RDS protocols vying for adoption in the U.S.: Sub-Carrier Traffic Information Channel (STIC), developed specifically for use in Intelligent Transportation Systems; and Data Audio Radio Channel (DARC). STIC receivers are expensive and require more power, but enable slightly higher data rates, 18 kbps versus DARC’s 16 kbps.
4.1.2 Implementation issues It is clear from the discussion above that there are numerous considerations in selecting a
become notorious for their use in increasing the odds of winning in various forms of gambling. It is possible, for example, to input data by pressing sensitive pads with the toes, and to receive output through miniature devices concealed in eyeglasses. Several gigabytes of storage can be concealed in clothing, along with powerful processing units. Of particular relevance to field GIS are devices that make it possible to see visual data displayed heads up in a unit that resembles a heavy pair of eyeglasses, and combines actual and virtual views. These devices are now used routinely in assembly plants, since they allow their users to see blueprints while working on assemblies. They could be used in field GIS to provide visual access to base maps, images, maps showing the landscape as it existed at some earlier time, or simulations of future landscapes. Systems such as these that augment reality are similar in some respects to immersive technologies, which replace reality with virtual renderings (e.g., Earnshaw et al. , 1995).
Goodchild (1998) argues that GIS should be seen as an interaction not only between human and computer, but between human, computer, and geographic reality (HCRI rather than HCI). A GIS database cannot be a perfect rendition of reality, since reality is infinitely complex but the database must always be finite, and thus some degree of approximation, generalization, or abstraction will always be necessary. Given this, effective GIS will always require interaction with the reality that the database imperfectly represents. Field GIS is an instance of HCRI, in which human and computer interact in the presence of the phenomenon, allowing the human to interact with reality while interacting with the computer. Field GIS allows direct access to ground truth, as well as to any digital representations of existing and prior conditions. As such, it promises to vastly improve the effectiveness of field research, and to open much more efficient channels of communication between field workers and the eventual users of their data.
4.3 Distributed computing
The networking functions provided by the Internet and its services support a wide range of modes of interaction between computers (for a review of GIS applications of the Internet see Plewe, 1997). But these are largely limited today to binary interactions between two computers. In a client-server environment, operations are divided between the client, which is commonly a machine of limited processing, storage, and software capabilities, and a server, which may have access to large resources of data, and much more powerful software and processing. Suppose, however, that a GIS user wishes to obtain two sets of data for the same area, and these are located on different servers. It is possible to download one set of data from one server, disconnect, and reconnect to the second server to download its data. It is not possible, however, to access both servers simultaneously, or to make use of services that avoid download of data. If a map is needed, it must be computed at the client from two downloaded data sets. This is in contrast to a map based on a single data set, which can be computed at the data set's server, avoiding the need to download data or to have mapping software at the client.
Thus while the Internet and WWW offer many powerful functions, they fail at this time to support many important but advanced capabilities that are part of the vision of truly distributed computing:
- simultaneous access to multiple servers from a single client;
- support for true distributed databases, such that the user sees a single database, but tables or even parts of tables are resident at different server sites;
- support for truly distributed software, such that the user sees a single software environment but modules remain resident at different server sites.
In all three of these areas there are active research projects and prototypes, but no comprehensive solution yet exists.
The WWW Mapping Special Interest Group of the Open GIS Consortium ( http://www.opengis.org/wwwmap/index.htm ) seeks to develop prototype demonstrations of solutions to the first bullet above, for GIS applications. Its aim is to make it possible for a client to access layers of data resident on multiple servers, while allowing the user to work with them as if they were all resident locally. For example, a user should be able to display and analyze a layer of soils and a layer of roads as if they were in the user's own GIS, without downloading them from their respective servers. Many problems will have to be overcome, including specification and adoption of standards, before this vision becomes a practical reality, despite the fact that it is already reality in the case of data on a single server.
In the case of the second bullet, there is much interest in the GIS community in making it possible for different agencies to own and maintain different parts of a single, unified database. For example, responsibility for the fields in a streets database might be divided between a mapping agency, with responsibility for defining and maintaining the basic geometric and topological framework of streets; and a transportation agency responsible for adding dynamic information on levels of congestion and construction projects. Since the latter will be represented as attributes in tables defined by the former, it is clearly preferable that parts of the database exist on different servers. But there are numerous problems in developing the technology of distributed databases (Burleson, 1994). Maintenance of integrity is particularly difficult, if different users are to be allowed to access and modify parts of the same tables simultaneously. An open question is the extent to which distributed GIS databases will be based on generic distributed database technology, or supported using conventional technology through administrative arrangements and protocols.
In the case of the third bullet, much effort is currently under way in the GIS software industry to exploit modern capabilities for distributed components. The underlying standards and protocols are being developed by the Open GIS Consortium ( http://www.opengis.org ), a group of companies, agencies, and universities dedicated to greater interoperability in GIS. Already it is possible to integrate GIS and other software environments, by allowing:
- a user in a GIS environment to access directly the services of some other environment, provided both are resident locally;
- the reverse, for a user in a non-GIS environment, such as a spreadsheet, to access GIS functions without leaving the current environment;
- a user to combine the services of a remote host by integrating them with the services of a local client.
5. AN ECONOMIC FRAMEWORK
The previous section on locational history has already hinted at how the location of computing might be placed within an economic framework. This section expands on that theme, and presents a basis for research on the costs, benefits, and economic value of distributed and mobile computing.
From the communications perspective established earlier, computing is seen as a process of information transfer from one person, group, or agency to another. Locations are associated with both sender and receiver, since the human intelligence associated with both must be located somewhere. Locations are also associated with storage, processing, and all of the other stages identified in Section 2.2.
5.1 Transport costs
Various costs are associated with the need to overcome the separation between the locations of sender and receiver, and with other separations such as that between the location of geographic ground truth and that of the sender. In classical location theory these costs are tangible, being determined by the costs of moving people or goods from place to place. In the case of information, however, there are both tangible costs, related to renting communication links and the fixed costs of establishing them, and intangible costs related to delay and unreliability.
Consider the costs associated with sending information between locations i and j. Several distinct conditions are relevant in today's environment:
- There exists a fixed communications link of high capacity, whose fixed costs have been absorbed by some agency and for which rental is free. Tangible costs are zero, as are intangible costs. This is the normal case when the Internet is accessible and both sender and receiver are in the U.S.
- There exists a fixed high-capacity communications link, but it has significant latency; while tangible costs are zero, there are substantial intangible costs to using the link.
- No fixed high-capacity communications link exists. Communications must rely on low-capacity links such as telephone lines, which may incur rental charges; or on wireless links with rental charges and substantial unreliability. There are both tangible and intangible costs associated with communication of information.
When costs exist, they begin to influence locational decisions, and either the sender or receiver may choose to relocate, other locations needed for the communication of information may be affected, or communication links may be chosen, to minimize transport costs.
5.2 Facility costs
Various facilities are needed for communication to occur between sender and receiver. Computer processing may be needed to modify or transform data, or to provide analysis or modeling. Processing will also be needed at the locations of servers, and at other nodes in the communications network. As in the previous section, many of these processing
resources are available free because they are part of the Internet, or provided through other arrangements to the sender or receiver. Others must be covered by the users through rental or fixed cost charges. Locational decisions may be involved if there is the possibility of selection among alternative processing locations, or alternative computing resources. Costs may be tangible, when processing power must be rented or purchased, but they may also be intangible, as when choices exist between the computing resources of different agencies, and issues such as security, privacy, or intellectual property are important.
5.3 Human intelligence as a locational factor
Finally, the locational decision will be influenced by the need to consider the locations of human actors in the system. If GIS is a communication problem, as suggested here, then the locations of sender and receiver are both important. Other human actors may also be involved, as interpreters, custodians of ancillary data, or developers of software. In Section 3 it was argued that in earlier phases of the history of computing it was common for human intelligence to move in response to the availability of computing resources. The decision to compute in the office rather than the field may also be an instance of moving human intelligence to the location of computing resources, rather than the reverse. As computing becomes cheaper and the costs of communication lower, it will be increasingly common to move computing to human intelligence, rather than the opposite; and arguably that process is already almost complete with respect to locations provided with power supplies and Internet connections. Changing economics of computing and emerging field technologies will have substantial influence on the locational decisions made by GIS users.
Locational decisions such as those discussed in this chapter will clearly impact where computing is done, and where users choose to locate, both for scientific and for decision- making applications of GIS. An important area of research is emerging, in the development of models and frameworks that allow such decisions and options to be explored in a rigorous, well-informed framework that can make use of our increasing understanding of the costs of computing and communications.
6. LIBRARIES AND ARCHIVES
Like any other institution, libraries are feeling the influence of the shift to digital communication, and the concept of a digital library has received much attention in recent years. In principle a digital library is usable entirely through the Internet, and thus achieves universal access. Its catalog is digital, and all of its books and other information resources are also in digital form. Novel forms of searching are possible; information can be sorted and delivered independently of the media on which it was originally stored; and information can be processed and analyzed either by the library or by its users. In short, the digital library holds enormous promise for increasing humanity’s access to its information resources. This section discusses several aspects of digital libraries of relevance to geographic information, in the context of distributed and mobile computing.
