Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 9
Toward A National Research Network 3 RESEARCHERS AND NETWORKS: APPLICATIONS AND USER CONCERNS JUSTIFICATION FOR A NATIONAL RESEARCH NETWORK The committee found that science researchers recognize a number of compelling reasons for an NRN that emanate from current as well as anticipated developments in the conduct of research. Among those that come from current user requirements and technology trends are the following: Researchers face a growing need for communication with professional associates, a need that electronic mail (e-mail) has begun to serve already. As discussed below, researchers have called for expanding the capability of e-mail services to include integration of text, graphics, animation, color, and speech. Overall, computer networking greatly enhances collaboration, avoiding problems due to multiple time zones and personal schedules or delivery delay that can make telephone and mail communication unwieldy. The advances in networks have not nearly kept up with the increase in the power of computation; the situation is analogous to limiting high-performance automobiles to footpaths. The capabilities implied by a high-performance network would help to make the most of existing and anticipated high-performance computing resources. Specifically, the advent of supercomputers has introduced capabilities for computation, simulation, modeling, and so forth, that generate vast data files that increasingly are represented graphically. These files cannot be communicated to associates in remote locations in reasonable time via current research networks.* Very large bandwidths are needed in certain applications. For example, to send (uncompressed) animated color graphics in real time, we can easily generate an individual bandwidth need for 0.75 Gbits/s as follows. Assume we have a * McCormick, Bruce H., et al., eds. 1987. Visualization in Scientific Computing, Report to the National Science Foundation by the Panel on Graphics, Image Processing and Workstations. Association for Computing Machinery Special Interest Group on Graphics, New York.
OCR for page 10
Toward A National Research Network 1024x1024 raster display (i.e., roughly 1 million picture elements, or pels), each pel requiring 8 bits to specify each of three colors (red, blue, green), and requiring 30 frames per second for real time. Then we need 1024 x 1024 x 3 x 8 x 30 bits/s, or roughly 750 Mbits/s on an individual basis. Multiplying this bandwidth by some estimate of the number of users who might simultaneously be transmitting such graphics over the same long-haul link easily leads to a requirement for gigabit per second links. Of course, one can often apply a significant amount of video-image compression to this kind of application, which could reduce the demand for raw bandwidth generated by this application in those cases where compression was used.* There are many large data-generation and observational efforts (for example, those undertaken by the National Weather Service, National Aeronautics and Space Administration, and National Oceanographic and Atmospheric Administration) that produce very large files at great expense to taxpayers. The data files are of enormous potential value; many researchers want to use them but cannot gain network access to these centrally located databases. Computational research may take place in distributed locations. Distributed systems require adequate network access to remote computers. There are further reasons for establishing an NRN that flow from the potential to improve the structure of the research community itself. Among them are the following: Because the number of universities housing first-class research facilities tends to grow far slower than the growth in the number of first-class researchers, many first-class researchers are dispersed among institutions with only limited capacity to support them. An NRN would enable those researchers to continue and strengthen their research by providing access to colleagues as well as better facilities. Access through an NRN would help to upgrade many types of research at a wide range of institutions more quickly and at a much lower cost than through such means as fundraising. A great benefit in terms of research quantity (and possibly quality) may arise by providing equitable access to those who are currently isolated by geographical location. This is illustrated by remarks from an oceanographer elicited by the committee when it sought input from the research community: * Whereas the effect on network response time of the propagation delay due to the finite speed of light is completely insignificant when transmitting (even for large files—say, 1 Mbit—and even over long distances—say 3000 miles) over low-speed lines (i.e., no faster than 1 Mbit/s), it is a dominating effect when such file lengths and distances are used with very-high-speed lines (at least 100 Mbits/s).
OCR for page 11
Toward A National Research Network Having used the OMNET, SPAN, NODS, Arpanet networks plus DECNET links to FSU, FNOC, and NRL, I feel we have developed an appreciation for the need and importance of such links for our research. Being in a rather isolated part of the country and not being on a university campus means we rely heavily on the networks to connect us to the rest of the research community and even to our sponsors. I am convinced that to ensure that the government labs and the universities work in a more coordinated way will require that the network links be expanded and improved. An NRN would help to rationalize the schedule for developing advanced and expensive research tools (future generations of supercomputers) since they can be shared as soon as they are developed, much as simulations that contribute to the design of those tools could be shared. Note, however, that an influx of users focused on research in a wide range of fields would increase the focus on good service, promoting standardization. These outcomes could potentially slow the introduction of new tools, a process that can be disruptive. The mobility of researchers and their ideas would increase substantially. The near-equivalent of face-to-face collaboration in the presence of advanced and scarce research tools could occur almost anywhere through networking. Finally, there is the straightforward economic benefit of reaping economies of scale in the purchase of network facilities, economies that can be passed on to researchers and research funders in the form of lower costs for service. STATUS QUO: INADEQUATE SUPPORT AND FRUSTRATION Currently available research networks are fragmented into many separate and underfunded regimes (see Figure 1). Communication options are limited in terms of the kinds of computer equipment, systems, and applications that can use the networks effectively. Often, existing networks are only weakly interconnected or not connected at all, limiting the extent to which colleagues can use the networks to communicate and collaborate. Communication is also restricted by severe overloading of the networks, which can slow down or even prevent access; networks commonly used by researchers can be unstable, unpredictable, and extremely frustrating to use. Further, in general, instruction, documentation, and troubleshooting support are limited, resulting in a situation where those able to make the most out of existing research networks are those who have acquired expertise in computer networking science and technology.* * McAdams, Alan K., et al.; 1987. Market and Economic Impact Study: Initial Marketing Analysis (Final Report), NYSERNET. McAdams, Alan K., et al.; 1988. Market and Economic Impact Study: Lessons from Four Networkers (Second Report), NYSERNET.
OCR for page 12
Toward A National Research Network These problems plague researchers seeking to use such networks as Arpanet even for such relatively undemanding applications as e-mail. They are even more troubling to researchers using the early NSFNET to access NSF-funded supercomputers. NSFNET limitations have not only frustrated researchers, as clearly reported to the committee, but also inhibit effective use of expensive supercomputing resources. (Plans to upgrade NSFNET, announced in the fall of 1987, may relieve some of these problems.) Networking problems that researchers faced stem in large part from inadequate funding support for networking. With inadequate funding support, researchers have few network options from which to choose. Further, the networks used by many researchers leave them without influence on the services that are provided. Communication budgets for the individual researcher (or research organization) are minimal or nonexistent. In some cases, the government, as a sponsor of research, absorbs the costs. In others, the researcher simply goes without adequate network services. Under these conditions, the U.S. research community appears to be working far below its potential; it is not able to achieve many of the benefits of research networking. Funding issues are explored in Chapter 4. GENERAL APPLICATIONS Science researchers currently use networks for a wide variety of functions: Electronic mail (e-mail) and electronic publishing. File transfer. Graphics and image file transfer. Remote computer access (interactive and batch). Remote access to computer data banks. These functions use the network strictly as a tool for some other research problem. Electronic mail is the most pervasive use of networks by the science research community at the moment. It is not an application that, as currently conducted, requires the advanced capabilities proposed for the NRN. However, researchers would particularly benefit from a more advanced form of e-mail that would be supported by the NRN: e-mail that combines text with graphics and even with dynamic imagery. The research community overall depends on the sharing of knowledge through publication of research results, and publication through journals typically combines text with figures of some sort. For example, the November 27, 1987, issue of Science (selected at random) has about 30 photographs or two-dimensional figures (excluding simple x-y graphs and bar charts) and eight color photographs in 72 pages of text. Many of the black-and-white photographs would be more informative if they were in color. At least one paper reproduced several frames from a movie.
OCR for page 13
Toward A National Research Network The availability of color computer graphics and animation-generation systems for personal computers combined with standard protocols for inclusion of images and movies in e-mail is likely to make e-mail as important a medium of communication as print now is. The advantages of instantaneous communication electronically over the months’ or years’ delay of printing and distribution systems for print, along with electronic means for broadcast and sorting, should be decisive. Current distribution systems for film are awkward and expensive compared to printed media, and so the ability to transmit color movies in the same electronic framework with text and figures is likely to lead to major increases in the use of color movies to transmit both experimental data and results of computer simulation. The committee believes it is reasonable to expect electronic transmission of text, images, and movies on the eventual NRN to be as commonplace as the distribution of scientific journals or communication between researchers by telephone now is, and that minimum bandwidth requirements for the national network could be estimated from these comparisons. Such capabilities would satisfy unmet needs of specific segments of the research community for imagery. Oceanographers, for example, need to transmit satellite images in digital form to allow remote researchers to work with them, something they are generally unable to do with today’s research networks. The discussion above about how message, image, and file transfer applications may develop builds on what we know about current messaging activity. However, the committee expects new applications to emerge as a result of the new capabilities that will be provided to the user. This is a critical point: we know from past experience with, first, early computer centers and, later, early research networks, that providing computer and communications resources has led to growth in demand far exceeding early forecasts. Further, we have seen an evolution in computer and communications applications that has been qualitatively as well as quantitatively different from what had been forecast. Electronic mail, for example, was not anticipated to dominate Arpanet traffic, but it did. Thus past experience shows that planning for an NRN will never have the benefit of precise specification of requirements. Indeed, probing may yield the wrong answers. The research community at large does not offer the well-defined user behavior assumed by planners of conventional data communications networks. The committee’s forecast of escalating requirements for network capabilities comes in no small way from the phenomenal growth in computing capabilities and applications. More powerful data processing equipment feeds the growth of more powerful data communications capability (at the local network level as well as at the wide-area network level). As computing technology progresses, two changes will occur. First, the progress will affect our ability to create demands on network performance in a very positive way. Second, researchers will conceive of new tasks that will lead to new demands on network performance. Indeed, while individual researchers will have far greater local computing power on their desks, their desktop power will remain considerably inferior to what can be amassed (at great expense) at a few central locations; access to these facilities will require improved networking capability.
OCR for page 14
Toward A National Research Network Computational researchers in many fields can today define the needs for large-scale computing many orders of magnitude more powerful than that which is available today. For example, computational chemists now struggle to model relatively small biological molecules. Consider not only modeling larger (by a factor of 100 to 1000) molecules, but also doing parameter studies of potential interactions between large numbers of pairs of such molecules. Consider modeling an assembly of such molecules (a cell). Consider modeling a simple organism or a complex organism. Users will come to expect strong local processing capability as well as seamless access to high-powered remote processing engines and database machines. The conclusion is that user desires are likely always to outstrip the capability of any available network, just as what is meant by supercomputers will continuously change with the state of the art. The meaning of large database will also change, and so it may always be necessary to ship the largest data files via postal-type service. However, the NRN has the potential to leapfrog current technologies sufficiently that we will meet the user needs well into the future, also providing a platform from which we can accelerate further capabilities well ahead of the rest of the world. Resource sharing is important in a wide variety of disciplines and will become more so as existing networks expand and are enhanced. It will be used in more disciplines as new tools and applications are developed. The ability to share resources facilitates collaboration of experts in joint research efforts and permits researchers to address large problems and complex systems that could not have been handled by a single expert before the existence of adequate networking for such sharing. Remote access in resource sharing is both batch and interactive. It includes what may be characterized as person-to-computer remote access and computer-to-computer remote access. Note that realizing the benefits of resource sharing would require not only appropriate network connections, but also such institutional mechanisms as recovery of royalty fees. In most disciplines where computers are used, it is common to use standard software packages for certain classes of problems. These packages may be produced by commercial entities (the third-party software industry) or by individual researchers or groups of researchers. Network access to remote computers facilitates the use of such software. The benefits of such sharing include cost reduction (some such software packages are very expensive) and the elimination of extraneous sources of difference when comparing two researchers’ results. Some of these software packages are difficult and/or time-consuming to implement and maintain on a given system, or may not even be available for a particular computer. Thus network access to remote facilities becomes crucial to certain types of research. Examples of program codes that have one or more of the above characteristics include NASTRAN and ANSYS in mechanical engineering, Gaussian86 in chemistry, MCNP in nuclear engineering, and SAAM in medical research. Most researchers prefer to use existing code to do research in chemistry and medicine, for example, rather than writing new code, because writing code is a large undertaking, and the progress of research would be severely impeded if
OCR for page 15
Toward A National Research Network most researchers had to develop their own code. As a corollary, the authors of widely used codes, such as Gaussian86 and the NIH SAAM system, often enhance these tools, and most researchers want access to the enhancements as soon as they become available. This process is greatly speeded up if the number of copies of the code is relatively small, which can only be the case if the code is readily available via remote access. Another type of facility that requires remote access for maximum usefulness is the large database. Examples of these abound and are growing rapidly as progress in the ability to collect and store large volumes of data grows. In nuclear engineering and in low-energy physics the Evaluated Nuclear Data File (ENDF) has long been a standard. The chemistry and medical research communities now have several such databases, including GENBANK and the Cambridge Crystallographic Database. Newly emerging databases include the human genome sequence and the global seismic data bank. These databases generally have two important characteristics in common for this discussion. First, these databases are accepted standards in their respective communities. Research using different values of similar data is more acceptable than research using an arbitrary value for the speed of light or mass of an electron. Second, the databases are quite large and becoming larger; the limit appears to be the ability of technology to provide large quantities of rapidly accessible storage. It is not practical to have many copies of such databases, any more than it is practical to have a full library of books in every researcher’s office. Most researchers only need some portion of these large databases at any given time. Few, if any, researchers will ever need access to all of a given database, any more than they would require access to all of the books in the library. The maintenance and updating of many copies of these very large data bases is horribly complex and surely should be avoided. Again, as in the case of the software packages, some of these databases are provided commercially and some are provided by individuals or groups of researchers. There are costs of acquisition and maintenance of these large databases. These costs transcend the fees and include the time and effort required to make them available. Centralizing this effort frees time for individuals to devote to their actual research. We can also expect to see increasing requirements for the use of “certified” data in certain engineering design applications. This too leads to a need for remote access to databases through networks. Additionally, access to major, large central stores of data as well as access to specialized files and databases of experts (including numeric and image results of experimental runs and numerical models) permits data exchange, thereby hastening the pace of research. Yet another facility that produces a need for remote access via a network is the specialized device. This may be a very expensive device such as a supercomputer that cannot be economically justified at every user site, or a new and experimental device that simply cannot be provided at more than a few sites because in the early stages of development only a few examples actually exist. High-energy physics, computational chemistry, mechanical engineering, atmospheric science, oceanography, astrophysics, nuclear engineering, and many other disciplines need access to supercomputers to solve leading edge problems. Some
OCR for page 16
Toward A National Research Network researchers in these disciplines also need to have access to machines with new architecture (and/or specialized architecture) to evaluate and utilize their particular characteristics. Not only would an NRN permit these researchers to gain access to the appropriate specialized devices, but it would also expand the availability of supercomputers to a wider user community. Also, an NRN with sufficiently large bandwidth would permit interactive use of supercomputers, which is precluded by the bandwidth and traffic limitations of current networks as well as by current job control procedures for access to some of these machines. Current developments in large-scale scientific computing are leading to truly distributed computing, where a given job is executed on several different machines as appropriate. The simplest example is the workstation front-end for a supercomputer where the user interface to code is presented by the workstation for both setup of the input data and analysis of the results, while the bulk of the computations are carried out by the supercomputer. The two machines may be located close together or they may be quite a distance apart. Such tasks require the simultaneous transmission of interactive and batch messages for a single user, not to mention for multiple users sharing a network. Other users may have primarily batch or primarily interactive requirements at any given time. It is absolutely necessary that the network be capable of providing rapid perceived response for interactive tasks in the presence of large file transfers. Large file transfer is a class of network usage that itself can justify an NRN. These file transfers may originate from large scientific experiments where the experimental device and the facilities used for analysis of the data are remote from each other. This can be the case in disciplines such as plasma physics, high-energy physics, and nuclear engineering. It will likely become the case in the future for such diverse disciplines as economics and clinical medicine. Another source of large file transfer requirements is remote sensing. The typical case is satellite imagery, but there are others (including, for example, the global seismic database is created from many diverse sensors on the surface of the globe). Some of these generate only a few bytes, others generate billions of bytes. Researchers working with image files have particular needs to transfer very large files in a speedy, dependable, and accurate manner. Currently, many of the networks have implemented forms of binary file transfer, but many technical problems make this file transfer difficult. One can still get a much more accurate transfer and a corresponding wider bandwidth by putting a tape or optical platter into a courier system for a one-day delivery. In order to satisfy the needs of the remote user, large file transfer must be made smooth and reliable. This must be combined with the needs for e-mail and other communications activities because the people who are interested in large file transfer are also very heavy e-mail users.
OCR for page 17
Toward A National Research Network DISCIPLINE-SPECIFIC APPLICATIONS The Utility Subcommittee of the National Research Network Review Committee sought examples (which are listed below) of network use (past, present, and future) from researchers in various disciplines. These examples are provided in order to demonstrate the needs and potentials of the NRN. Examples are the result of the interviews conducted by members of the Utility Subcommittee as well as responses to an electronic bulletin board solicitation. The large number of responses, along with the great length of some of them, clearly demonstrated the strong interest in the networks and their future on the part of the science community. Many people see the networks as a critical part of their research activities and want to ensure at least their present level of access as a bare minimum, and they stress the need for great improvements in the future. A selection of anecdotal reports (wants, needs, uses, expectations, complaints, and general observations) on network use by researchers is provided in this section. While this survey is neither complete nor systematic (the committee felt that the time required for a complete systematic survey was incompatible with the pressing need for action), it does illustrate the breadth of current and potential applications for an NRN in science/engineering, economics/social sciences, and arts/humanities. All potential users consulted agree that an NRN is a most worthwhile goal. Mathematics Both simulation and image generation are growing in importance in mathematics. Fractals and Mandelbrot sets provide a new framework for geometrical studies; chaotic dynamical systems create geometrical structures such as “strange attractors,” which require extensive simulation to be defined rigorously. Studies of partial differential equations, stochastic processes, and probability theory often require state-of-the-art supercomputing. The newly discovered overlap of gauge theories of elementary particles with advanced topological problems in mathematics is one of many areas with needs for symbolic algebra computations. The proof of the four-color theorem ushered in a new paradigm for proofs: the use of computers for combinational and algebraic theorems. Physics The needs of solid-state physicists are similar to those of chemists—to communicate three-dimensional electron density maps of unit cells of solids and solid surfaces and to have access to supercomputers for simulations of structure at zero and finite temperatures. Instrumentation such as electron microscopes, synchrotron radiation facilities, scanning tunneling microscopes, and so on, will increasingly generate complex images and movies requiring both supercomputing for analysis and wide bandwidth networks for communicating collision data, which already can run to thousands of tapes for a single experiment. Physicists need
OCR for page 18
Toward A National Research Network supercomputer access for simulations of quarks and other phenomena and for simulations of beams in particle accelerators. Physicists engaged in studies of fluid dynamics and turbulence need access to supercomputers for three-dimensional simulations of turbulent fluid flows and need to communicate three-dimensional images of these flows interactively while the simulations take place, so that they can understand these simulations and rework them when they go astray. Chemistry Throughout all of chemistry, chemists need to communicate three-dimensional chemical structures, both ball and stick displays and full three-dimensional electron density maps. Both electronic literature searches and e-mail need this communication capability. The electron density maps will require major bandwidth increases. Ultimately, movies of density changes during chemical reactions will be of major importance too. Chemists need remote supercomputer access for the generation of electron density maps by chemical modeling. The simulation of molecular structure, chemical reactions, and statistical mechanics of solids, liquids, and gases on supercomputers are all growing areas that, because of major computing demands, will require state-of-the-art supercomputing for the foreseeable future. Interactive image transmission to monitor long simulations (for example, of chemical reactions) will be of major importance. Astronomy and Astrophysics Astronomers and astrophysicists are becoming heavily dependent on images and image generation from ground-based facilities (optical telescopes, radio telescopes, and so forth); space flights of all kinds (space telescopes, planetary flybys, and so on); and computer simulations. Computer simulations are and will be generating three-dimensional images of planetary atmospheres and rings, the interiors of stars, entire galaxies, black holes, and quasars, and simulations will be developed all the way to large-scale structures of the universe itself. As computer simulations become three-dimensional, supercomputer access and interactive transmission of images are required for future progress.
OCR for page 19
Toward A National Research Network Atmospheric Sciences Including Oceanography As with most disciplines, the primary use of networks by atmospheric scientists is for e-mail.* Notably, the transfer of large binary files is often an important activity. Large files must be transferred not only between supercomputers but also between smallermachines and workstations. Line graphics and images are important components of atmospheric science research and must also be transferred via networks. Another rather conventional but equally important use of networks is by atmospheric scientists using remote access to large computing facilities such as supercomputers. Supercomputer users are working throughout the United States. The use of supercomputers, for modeling and/or statistical analyses, requires connection to networks. This use even extends to ships at sea and to other scientists in the field (for example, Antarctica). Within atmospheric sciences, as well as other disciplines, there is a need for online access to databases. Currently, when a researcher requires a data set from a remote network location, the most common way to obtain it is to request it in the form of a tape, and it may take weeks or even months to receive it. Some of these data sets are not extremely large and could easily be accessed remotely. Remote access provides an added advantage in that the investigator who may want only a small portion of the data file can select only the desired part rather than having to order the entire data set, most of which he will not use. Online access becomes more complex when one considers satellite image files that are much larger than other data files. Still, it is realistic to first be able to query the library index file to determine just what data are available and then, for limited requests, send image files across the network. This is a critical development that is taking place now, but such transfers are currently fraught with difficulties. NASA has developed a variety of online access systems such as the NASA Ocean Data System (NODS), which is a catalog of satellite data and a repository for a limited amount of data. Plans are to expand NODS to * It is interesting, however, that within this group the need for electronic communication was so great that a separate e-mail network was set up using the commercial Telenet system sold by Sprint Communications. Called OMNET, the value-added resale venture is run by an oceanographer and his wife and has now expanded well beyond the atmospheric science community into a large number of other scientific disciplines. Their service is called Sciencenet. It includes about 2000 users. It supports personal communications as well as bulletin board announcements, binary file transfer, and some graphics transfer. This system is effective because the OMNET management carefully introduces new users to the system and gives users continuous updates regarding the system as well as likely changes. OMNET publishes a newsletter and continuously seeks new ways of serving users. Thus users find that the value-added service of Sciencenet is worth the small added cost of using it as compared to using Telenet directly.
OCR for page 20
Toward A National Research Network contain more data and to use remote data archives for data access. NASA has developed the Global Ocean Library Data (GOLD) catalog system for the remote access to future satellite data archives. Other NASA systems collect and maintain land data, climate data, and planetary data. All are intended to be made accessible online. There are also plans for an online data system as part of the World Ocean Circulation Experiment that will commence in 1990. This type of access is the wave of the future and is eminently sensible in terms of providing needed data to the user in the quickest possible fashion. The current tape delivery system for meteorology and oceanography is inadequate, cumbersome, and frustrating. See Appendix B for an expanded examination of oceanographers’ needs, provided as a detailed illustration of scientists’ networking needs. Engineering Current applications in engineering include supercomputer access for calculating the stresses imposed on skyscrapers in the vicinity of earthquakes. Such analyses must be interactive. Indeed, the power of interactive simulation is such that simulations done in batch mode are almost extinct. This assertion covers fields from circuit design to integrative biology. To the extent that large-scale simulations will require access to remote machines, a network capable of supporting interactive remote logons will soon be essential to progress in these fields. Researchers in mechanical engineering need supercomputer access and communication of three-dimensional images in major areas such as aircraft simulation, vehicle simulations (for air resistance, crashworthiness, and structural analysis), turbulent fluid simulations (coal slurry transport, mineral and powder processing, chemical reactor flows), simulations of noise generation by vehicles and aircraft, simulations of the particle mechanics of soil, simulations of metal fatigue, fluid simulations of oiled bearings, and so on. The primary network needs for structural engineers are access to large computing resources, such as supercomputers, and remote access to both digital and graphic results. Most structural design is carried out using large programs with many options, and the interpretation is then done on personal workstations. Thus the networks must be able to execute the programs on remote machines and then transfer the results to the remote user workstation. Until recently, the results were analyzed in numerical form, but now they are increasingly analyzed in graphic form on personal workstations. These generalities apply to various parts of structural engineering ranging, for example, from the design of buildings and bridges to the design of the Space Station. In this design process, remote users must access computational results produced by prime contractors who carry out most of the computational experiments but rely on a wide variety of users to interpret the results. Thus many remote users need dependable links to Boeing and Lockheed to participate in Space Station design and evaluation.
OCR for page 21
Toward A National Research Network The network needs of electrical and computer engineers are of two types-research needs that relate to the design and development of networks and those that do not. In the first case, communications engineers are directly involved with the network as well as being dependent on the network for support facilities. Electronic mail is important, as is the exchanging of test ideas for communications research. Computers, and hence, networks are also used for simulations of complex communications systems, and the network may be used to transfer the simulation results that determine its own future. Whereas the use of a network as a test bed may suffice in a university research environment, it would not be advisable in a service environment, which is what users want the NRN to be. The second type of network use in electrical engineering involves research topics that do not directly relate to network design and development. The network serves a support role as is the case in most other disciplines. Remote access to computer resources is required, as are the transfer of simulation study results and the use of e-mail for communications. Application examples include large-scale VLSI design, simulation of distributed systems, large-scale programming efforts, modeling and simulation of multipath-multihop communication systems, and signal processing of video and speech data. Researchers in chemical engineering need supercomputer access and communication of three-dimensional images for simulations of thermodynamic properties of liquids and gases; for molecular simulations of various kinds; for simulations of fluid flows, chemical reactions, and flames in the presence of turbulence; for simulations of fluid flows in pores and porous media; for simulations of catalysis of all kinds, and for simulations of interactions of gases and liquids on catalytic surfaces. Life Sciences and Medicine Medical researchers are active network users. The generation and transmission of images are becoming a key need in the life sciences and medicine. Three-dimensional images of living units, from individual biological molecules (proteins, DNA, and so on), to cells, to major parts of bodies of animals and humans, are becoming of central importance as scientists learn how to generate and manipulate these images. Supercomputer access is needed for simulations and image analysis, and instantaneous image transmission is needed for communication of these images from sources such as a patient in a hospital, a large synchrotron x-ray facility, or a centrally managed database to the expert or researcher who will try to understand the image. Simulations in biology and medicine require state-of-the-art supercomputer access for studies ranging from molecular structure (protein folding and other structural information that is at the basis of many biological processes, including drug action) to simulations of neural networks and brain functions. Being interested in large-scale simulation and modeling of human, animal, and plant physiology, the integrative and computational biology communities are particularly sensitive to the need for a more extensive research network. This group aims at integrating the results of many laboratories into consistent theories describing intact organisms. Because it is no longer possible for a
OCR for page 22
Toward A National Research Network single laboratory to make all the measurements that are important for a given area of biological research, close electronic ties are essential to any effort at integrating the accumulating mass of experimental information into a useful, predictive theory. The National Institutes of Health, Division of Research Resources, has established several biological simulation resources aimed at bringing this technology to the biomedical community. As yet, however, these resources are not linked to major electronic databases; they are not even linked to each other. Many medical researchers see the main advantage of networking to be the simultaneous availability of disparate pieces of information on a single display. Some of the nation’s leading radiology departments have committed to digital transfer of radiographic images on broadband local networks operating at 100 to 1000 Mbits/s. A national network capable of sufficient bandwidth would make it possible for radiographic images generated at small research hospitals to be read by specialists at distant locations whenever desirable. Medical researchers, like researchers in other disciplines, use networks for sending abstracts, data, and manuscripts to collaborators around the world. Connections are made, for example, to the Technion in Israel, Rotterdam in the Netherlands, Auckland in New Zealand, as well as to other research centers in the United States. Network users are motivated by speed and convenience. For example, questions about clinical experience in Israel can be asked and answered in one day; the asking of such questions would probably never be considered in the absence of the network. This group, however, has not yet been able to transmit static three-dimensional graphic images representing the solutions of mathematical models of cardiac function. Economics and Social Sciences A major challenge to economic and social sciences is to predict economic and social behavior of bulk populations and entire nations in the presence of random actions of individuals. Access to supercomputers is required for simulations of large populations covering the full variance of each random process. Interactive multidimensional displays from these simulations will be as demanding of network bandwidth as in the hard sciences; for example, simulations of international trade negotiations involve four-dimensional displays just for bilateral trade data (nation in versus nation out versus commodity versus time gives four dimensions). Other demanding areas are simulations of occupational and demographic flows and simulations involving combinatorial optimization in a stochastic environment, such as simulations of traffic flows on city streets to optimize traffic light settings and the like. Access to economic and social databases is a major need, although high-bandwidth access is not required as long as these data come from surveys and not from simulations or other computer-generated sources. Note that database material in the social sciences may be more combinable than in the physical sciences.
OCR for page 23
Toward A National Research Network Arts and Humanities Perhaps surprisingly, there are network applications being developed in departments that might be seen as the very antithesis of modern technology. For example, a lexicon of ancient Aramaic languages is being developed in the Department of Near Eastern Studies of a major university and the chairman of the Classics Department in the same university has constructed a 300-Mbits thesaurus of ancient Greek literature. The importance of the network to these specialized applications is that the small community of scholars interested in these databases can work synergistically even though based at geographically separate institutions. Moreover, they can do so without the inefficiency of replicated computational resources. USER CONCERNS While researchers generally appreciate what computer networks may do for them, they have concerns arising from recent experiences with today’s networks and from expectations about how an NRN might be implemented. Researchers are not a homogeneous community with uniform requirements for network support. While the NRN would provide the research community overall with vastly enhanced capabilities, some of the needs envisioned by researchers today would not be met by the proposed NRN. This is primarily because it may not be economically feasible to satisfy all of the needs of a heterogeneous research community. The committee has not identified any capabilities that appear to present insurmountable technical problems. Nevertheless, the government must avoid the danger of making unrealistic promises to the research community. Part of the planning process should therefore include a delineation of those applications that can be supported (and how) and those that cannot. Researchers are generally confused by the current state of network systems with which they must deal. As a result, they are extremely concerned that an NRN not be used as a test bed for new networking techniques and technology; they require an operational, functioning, user-friendly, reliable service network. Users will see the network as consisting of all components between themselves and the object of their communication. If there is a broken component, users cannot and should not be expected to troubleshoot and identify the problem. Use of a network as a test bed might be acceptable in a communications research environment but not in a service environment. There is a considerable concern within the research community as to whether or not an NRN could be effectively managed. As discussed below, management affects access, user support, and funding as well as introduction of new capabilities and services. Given their frustration with the status quo, researchers need and want a system that demonstrates responsiveness to their concerns. The potential of an NRN would not be realized, and it would be a costly mistake, if the government promoted a network with a large number of displeased users.
OCR for page 24
Toward A National Research Network Researchers need and want to be informed of changes (good and bad) on a regular basis and have input into major changes. Users need, and want, to be protected from changes in protocols and procedures-the emphasis should be on providing access to a valuable research tool, not on converting all researchers into network gurus. Users require a transparent network system through which they can communicate with a variety of attached devices and protocols; they should not need to worry about the incompatibilities between TCP/IP and DECNET systems, for example. This is not to say that researchers are lobbying for standards, but rather that they seek a pervasive level of connectivity. Such connectivity is required to achieve the collaboration and communication potential promised by an NRN; the world in which researchers, like others, operate is one marked by a very wide range of computer systems. Fairness to users in the management and operation of the network is a major concern for a network with the goal of providing broad access. If access is granted on the basis of ability to pay, the network may foster a “rich-get-richer” outcome. The problems faced by smaller research institutions are illustrated by remarks from another scientist: I would recommend that you address the needs of the [smaller] southern U.S. colleges and universities. Their needs are driven by: inadequate computational facilities to support a viable science-oriented research program; inadequate resources to buy outright their own computers/processors; a generally diffuse organizational structure (decentralized as opposed to the northeast universities where you can concentrate your computer centers); a historical pattern of non-participation in national programs…; an emerging technology-oriented industrialization in states where agriculture and forest products have long dominated; a competition for scarce educational resources… Equity and access to the small users are important considerations. They may be difficult to achieve in that extending service to areas of low population and user density may be relatively costly. On the other hand, a situation where large or “rich” users provide their own individual solutions to networking would prevent an NRN from achieving the benefits of scale or access promised by the concept of a truly national network.
Representative terms from entire chapter: