United States General Accounting Office ____________________________________________________________________________ GAO Report to Congressional Requesters ____________________________________________________________________________ July 1991 HIGH-PERFORMANCE COMPUTING Industry Uses of Supercomputers and High-Speed Networks Note: This report uses the # character to represent a fat line over the top of major headings. The upper ASCII 220 character would draw a better fat line that would be more like the line in the printed report. You can use your word processor to search for 21 #'s and replace them with 21 ASCII 220 characters. If you change the #'s your printer must be able to print the ASCII 220 character. This report represents an effort to make it appear as close to the printed version as ASCII will allow. To obtain a printed copy of this report call call GAO report distribution at 202/275-6241 (7:30 a.m.-5:30 p.m. EST) or write to GAO, P.O. Box 6015, Gaithersburg, MD 20877. ____________________________________________________________________________ GAO/IMTEC-91-58 This U.S. General Accounting Office (GAO) report is available over the Internet as part of a test to determine whether there is sufficient interest within this community to warrant making all GAO reports available over the Internet. The file REPORTS in the Anonymous FTP directory GAO.REPORTS at NIH lists additional reports; the file A-LIST contains a list of reports issued from September 1991. So that we can keep a count of report recipients, and your reaction, please send an E-Mail message to KH3@CU.NIH.GOV and include, along with your E-Mail address, the following information: 1) Your organization. 2) Your position/title and name (optional). 3) The title/report number of the above reports you have retrieved electronically or ordered by mail or phone. 4) Whether you have ever obtained a GAO report before. 5) Whether you have copied a report onto another bulletin board--if so, which report and bulletin board. 6) Other GAO report subjects you would be interested in. GAO's reports cover a broad range of subjects such as major weapons systems, energy, financial institutions, and pollution control. 7) Any additional comments or suggestions. Thank you for your time. Sincerely, Jack L. Brock, Jr. Director, Government Information and Financial Management Issues Information Management and Technology Division i _____________________________________________________________________________ GAO United States General Accounting Office Washington, D.C. 20548 ______________________________________________________ Information Management and Technology Division B-244488 July 30, 1991 The Honorable Ernest F. Hollings Chairman, Senate Committee on Commerce, Science, and Transportation The Honorable Al Gore Chairman, Subcommittee on Science, Technology, and Space Senate Committee on Commerce, Science, and Transportation The Honorable George E. Brown, Jr. Chairman, House Committee on Science, Space, and Technology The Honorable Robert S. Walker Ranking Minority Member House Committee on Science, Space, and Technology The Honorable Tim Valentine Chairman, Subcommittee on Technology and Competitiveness House Committee on Science, Space, and Technology The Honorable Tom Lewis Ranking Minority Member Subcommittee on Technology and Competitiveness House Committee on Science, Space, and Technology This report responds to your October 2, 1990, and March 11, 1991, requests for information on supercomputers and high-speed networks. You specifically asked that we - provide examples of how various industries are using supercomputers to improve products, reduce costs, save time, and provide other benefits; - identify barriers preventing the increased use of supercomputers; and Page 1 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks _____________________________________________________________________________ B-244488 _____________________________________________________________________________ - provide examples of how certain industries are using and benefitting from high-speed networks. As agreed with the Senate Committee on Commerce, Science, and Transportation, and Subcommittee on Science, Technology, and Space, our review of supercomputers examined five industries--oil, aerospace, automobile, and chemical and pharmaceutical. These industries are known for using supercomputers to solve complex problems for which solutions might otherwise be unattainable. Appendixes II through V provide detailed accounts--drawn from 24 companies contacted within these industries--of how supercomputers are used to improve products and provide other benefits. We also obtained information from 10 companies in the oil, automobile, and computer industries concerning how they are using and benefitting from high-speed computer networks. We did not verify the accuracy of the examples of benefits provided by the various companies. Appendixes I, VI, and VII, respectively, provide additional information on the objectives, scope, and methodology of our review, and identify the companies examined to assess uses of supercomputers and high-speed networks. #####################________________________________________________________ RESULTS IN BRIEF Supercomputers contribute significantly to the oil, automobile, aerospace, and chemical and pharmaceutical industries' ability to solve complex problems. They enable companies within these industries to design new and better products in less time, and to simulate product tests that would have been impossible without spending months developing and experimenting with expensive product models. Some companies have attributed significant cost savings to the use of supercomputers. For example, although exact figures were not always available, representatives of some automobile and aerospace companies estimated that millions of dollars have been saved on specific models or vehicle parts because of reduced manufacturing or testing costs. In addition, one oil company representative estimated that over the last 10 years, supercomputer use has resulted in increased production of oil worth between $5 billion and $10 billion from two of the largest U.S. oil fields. Despite widespread use of supercomputers for certain applications, representatives of these companies told us that several key barriers currently hinder their greater use. These barriers include (1) the high cost of supercomputers, (2) a lack of application software, (3) the cultural resistance to the shift from physical experiments to an increased reliance on Page 2 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks _____________________________________________________________________________ B-244488 _____________________________________________________________________________ computational experiments, and (4) a lack of supercomputing education and training. High-speed networks contribute to improved productivity by enabling industries to more efficiently share information and resources and collaborate on product development over distances. Companies within the oil, automobile, and computer industries, for example, rely on high-speed networks to transfer large graphics and data files, and access computers worldwide. In many cases these companies must use high-speed networks (as opposed to those operating at lower speeds) because (1) the applications in use require high transmission speeds, for example to provide instant images for interactive videoconferencing, (2) high volumes of traffic in one or more applications are being transmitted, or (3) fast response is needed for such applications as data base queries. Several companies reported that they would not be able to develop products in a timely manner without high-speed networks. #####################________________________________________________________ BACKGROUND A supercomputer, by its most basic definition, is the most powerful computer available at a given time. Current supercomputers, costing from about $1 million to $30 million, are capable of performing billions of calculations each second. Computations requiring hours or days on conventional computers may be accomplished in a few minutes or seconds on a supercomputer. Although the term supercomputer does not refer to a particular design or type of computer, supercomputers generally use vector or parallel processing. With vector processing, a supercomputer lines up billions of calculations and then uses one or several large processors to perform these calculations. In parallel processing, many smaller processors work on multiple parts of a program concurrently. The trend in supercomputer design is to add more processors to achieve greater performance. Massively parallel supercomputers consisting of between 1,000 and 64,000 processors now exist. The unique computational power of supercomputers makes it possible to solve critical scientific and engineering problems that cannot be dealt with satisfactorily by theoretical, analytical, or experimental means. Scientists and engineers in many fields--including aerospace, petroleum exploration, automobile design and testing, chemistry, materials science, and electronics--emphasize the value of supercomputers in solving complex problems. Much of this work involves the use of workstations for Page 3 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks _____________________________________________________________________________ B-244488 _____________________________________________________________________________ scientific visualization--a technique allowing researchers to convert masses of data into three- dimensional images of objects or systems under study. These images enable researchers to comprehend more readily what data reveal and facilitate the understanding of problems by different types of scientists and engineers. While supercomputers are still relatively limited in use, the number of supercomputers has risen dramatically in the last decade. In the early 1980s, most of the 20 to 30 supercomputers in existence were operated by government agencies for such purposes as weapons research and weather modeling. Today, about 280 supercomputers#1 are in use worldwide. The Government (including defense-related industry) remains the largest user, although private industry has been the fastest growing user segment for the past few years, and is projected to remain so. A high-speed network is generally defined as a network operating at speeds of T1--1.544 million bits per second--or higher. Prior to 1977, high-speed networks were employed exclusively by the telephone companies. By the early 1980s, however, these services had become widely available to commercial customers. Today, thousands of high-speed networks exist, fueled by demands for a variety of applications, such as electronic mail, data file transfer, and distributed data base access. Networks operating at T1-speeds are common and provide sufficient capability to meet most application needs. However, there is a growing demand for higher-speed networks, such as those operating at T3-speeds (45 million bits per second) or greater, to transmit multiple low-speed applications to many users at the same time. In addition, many industries now look to such networks as a means of transmitting more advanced applications that result from the use of supercomputers and other sophisticated technologies. The growth of T1 and T3 lines is expected to be great, according to Northern Business Information/Datapro, a research company and industry analyst, which projected that revenues for T1 and T3 will increase three-fold between 1990 and 1994. ______________________________________________________ 1 This figure includes only high-end supercomputers such as those manufactured by Cray Research, Inc. Including International Business Machines (IBM) mainframes with vector facilities would about double this number. Page 4 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks _____________________________________________________________________________ B-244488 _____________________________________________________________________________ #####################________________________________________________________ INDUSTRIES BENEFIT Supercomputers provide the five selected FROM SUPERCOMPUTING industries with the ability to develop new and better products more quickly. Although most companies within these industries could not provide precise figures to quantify the extent of gains realized, nearly all believed that supercomputers have enabled them to perform previously impossible tasks, or achieve significant cost reductions or time savings. Moreover, industry representatives believed that greater benefits would be realized in the future, as these companies move toward using more powerful supercomputers with thousands of processors. Of the 21 companies that commented on the issue, 19 said that they will be using massively parallel supercomputers to a greater extent in the future.#2 Details on each industry's use of supercomputers are in appendixes II through V. _____________________________________________________________________________ The Oil Industry As an early user of supercomputers, the oil industry has realized substantial benefits from supercomputer applications. By using two key applications for processing seismic data#3 and simulating reservoirs, oil companies have improved their ability to determine the location of reservoirs and to maximize recovery of oil and gas from those reservoirs. This ability has become increasingly important because of the low probability of discovering large oil fields in the continental U.S. New oil fields are often small and located in harsh environments, making exploration and production difficult. Several industry representatives estimated that the use of supercomputers reduces the number of dry wells drilled (at a cost of $.5 million to over $50 million per well) by about 10 percent. In addition, an Atlantic Richfield Company (ARCO) representative estimated that supercomputer use has led to increased oil production worth billions of dollars at two large fields. _____________________________________________________________________________ The Aerospace Engineers and researchers in the aerospace industry Industry have used supercomputers since the early 1980s to design, develop, and test aerospace vehicles and related components. Supercomputers, for example, have enabled engineers to analyze aircraft structural composition for design flaws and to simulate their ______________________________________________________ 2 Three out of 24 representatives did not comment on the issue for proprietary reasons. 3 Seismic data reveal characteristics about the earth and are gathered using sound recording devices to measure the speed that vibrations travel through the earth. Page 5 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks _____________________________________________________________________________ B-244488 _____________________________________________________________________________ performance in wind tunnels. This ability is important because wind tunnels are expensive to build and maintain, and cannot reliably detect certain airflow phenomena. Simulation permits a reduction in physical model testing, and substantial savings in time and money. As a major user of supercomputers, McDonnell Douglas estimates that supercomputer simulations saved about a year in the design and testing of its new C-17 military aircraft. _____________________________________________________________________________ The Automobile Since 1984, automobile manufacturers have Industry increasingly relied on supercomputers to design vehicles that are safer, lighter, more economical, and better built. By the late 1980s, the world's 12 largest automobile companies had acquired supercomputers. A primary supercomputer application-- crash analysis--is used to simulate how vehicle structures collapse on impact and how fast passengers move forward. These simulations provide more precise engineering information than was possible from physically crashing pre-prototype vehicles. They also reduce the number of vehicles required for these tests by about 20 to 30 percent. Consequently, companies have been able to save millions of dollars annually. According to General Motors Corporation representatives, for example, supercomputers enabled the company to crash 100 fewer vehicles when developing some of its 1992 models, than it did in 1987. Each test vehicle costs from $50,000 to $750,000, depending on whether a production vehicle or prototype is used. _____________________________________________________________________________ The Chemical and Supercomputers also play a growing role in the Pharmaceutical chemical and pharmaceutical industries, Industries although their use is still in its infancy. From computer-assisted molecular design to synthetic materials research, these companies increasingly rely on supercomputers to study critical design parameters and more quickly and accurately interpret and refine experimental results. Industry representatives told us that the use of supercomputers will result in new discoveries that otherwise may not have been possible. Du Pont, for example, is developing replacements for chlorofluorocarbons, compounds used as coolants for air conditioners, that are thought to contribute to the depletion of ozone in the atmosphere. In designing a new process to produce substitute compounds, Du Pont is using a supercomputer to make certain calculations needed for this process. Page 6 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks _____________________________________________________________________________ B-244488 _____________________________________________________________________________ These calculations, on a supercomputer, require a few days at a cost of between $2,000 to $5,000. Previously, however, such tests cost about $50,000 and required up to 3 months to conduct. #####################________________________________________________________ BARRIERS IMPEDE Although supercomputers have yielded highly GREATER USE OF visible contributions in the selected SUPERCOMPUTERS industries, representatives told us that many aspects of supercomputer use remain untapped, because of the following significant barriers. High cost: Currently, supercomputers cost between $1 million and $30 million, not including the cost of software development, maintenance, or trained staff. Lack of software: While the evolution of software for vector supercomputers has accelerated over the past decade, little reliable software has been developed for parallel supercomputers. This is in part due to the lack of software tools for developing new parallel software and converting vector software so that it can be used on massively parallel supercomputers. Cultural resistance: Many companies or industries, particularly the chemical and pharmaceutical industries, rely more heavily on physical experimentation than necessary, according to representatives. Many scientists and managers see the use of computational science as a dramatic break with past practice, and such a major shift in research methodology is difficult to accept. Lack of supercomputer training and education: Before 1985, university students and professors performed little of their research on supercomputers. Thus, for many years industry hired students from universities who did not bring supercomputing skills and experience to their jobs. According to Du Pont and Eli Lilly representatives, universities are still not providing a sufficient number of students skilled in the use of supercomputers. A Ford Motor Company representative also noted that there is a scarcity of trained staff in computational fluid dynamics, an important application to the automobile industry. Currently, formal supercomputer education is primarily limited to the National Science Foundation (NSF) university supercomputer centers. #####################________________________________________________________ INDUSTRY USES OF Like supercomputers, high-speed networks are HIGH-SPEED NETWORKS making valuable contributions to many industries. Companies in the oil, automobile, and computer industries, for example, increasingly rely on high- Page 7 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks _____________________________________________________________________________ B-244488 _____________________________________________________________________________ speed networks to share resources and provide various types of person-to-person communications. Many oil company representatives, in particular, reported network traffic increases, ranging from 10 to more than 100-fold over the past 5 years. Many companies thought that significant benefits--including monetary savings, reduced time-to-market, and improved product quality--have resulted from their use of high-speed networks. The companies we contacted primarily use high-speed networks operating at T1-speeds (1.544 million bits per second). A significantly smaller, although growing, number of companies also use higher-speed T3 networks of 45 million bits per second. These networks generally consist of private lines, leased exclusively for each company's use, although many are connected to outside commercial and private networks. Companies we contacted use high-speed networks for a variety of reasons. In some cases, these networks are used for individual applications that require high transmission speeds, such as interactive videoconferencing. Most companies also used these networks as a more cost-effective way of transmitting large volumes of aggregated traffic from lower-speed applications. These applications include voice communication, remote computer access, and electronic mail. Landmark Graphics Corporation, a company that develops seismic data processing software for oil exploration, for example, uses an extensive T1 network to support a variety of applications. This network supports up to four voice lines (at 64,000 bits per second each), while providing electronic mail access to hundreds of network users. This network also allows users across the country to work simultaneously on the development of the same software by accessing and sharing files via high performance workstations, and to routinely transfer voluminous files to backup the file system. A Landmark representative said that network use has provided more coordinated and consistent control of product development among the company's different offices, and ultimately, a shortened product development life-cycle. The Amoco Corporation uses a high-speed network to transmit very large (100 million bit to 1 billion bit) files between its foreign and domestic sites. The files contain large volumes of data such as images of sections of the earth, which measure about 400 square miles wide by 3 miles deep. These data are critical to improving Amoco's ability to locate oil reservoirs. Because of the volume of the data, Amoco Page 8 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks _____________________________________________________________________________ B-244488 _____________________________________________________________________________ representatives said it would be impossible to transmit these files to each work site without high- speed networks. If they did not have the networks, the data would have to be duplicated at each site, resulting in higher costs. Moreover, according to an Amoco representative, access to the supercomputer via the high-speed network enabled them to make a major oil discovery--the details of which are proprietary. Within the automobile industry, a General Motors (GM) Corporation representative reported that high-speed networks primarily benefit them by reducing costs and increasing productivity. For example, the network permits resource sharing, reducing duplicate hardware and software purchases. One group reported saving $90,000 by using university software over the network, rather than purchasing it. Another group reported that it did not have to buy a parallel supercomputer because it accessed one at a university via the network. In addition, a corporate networking group projected a $2.3 million cost avoidance for 1991 because the use of a high-speed network enabled them to make large data and graphics files more readily available to remote sites. _____________________________________________________________________________ We discussed the information in this report with industry representatives and experts, and incorporated their comments where appropriate. Our work was performed between October 1990 and May 1991. As agreed with your office, unless you publicly announce the contents of this report earlier, we plan no further distribution until 30 days from the date of this letter. We will then send copies to interested congressional committees and others upon request. Please contact me at (202) 275-3195 if you have any questions concerning this report. The major contributors to this report are listed in appendix VIII. Jack L. Brock, Jr. Director Government Information and Financial Management Page 9 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks _____________________________________________________________________________ CONTENTS #####################________________________________________________________ LETTER 1 #####################________________________________________________________ Appendix I 12 OBJECTIVES, SCOPE, AND METHODOLOGY #####################________________________________________________________ Appendix II 13 THE OIL INDUSTRY Seismic Data Processing 13 Reservoir Simulation 15 #####################________________________________________________________ Appendix III 17 THE AEROSPACE Computational Fluid Dynamics 18 INDUSTRY Structural Analysis 19 Computational Electromagnetics 20 #####################________________________________________________________ Appendix IV 21 THE AUTOMOBILE Automobile Crash Analysis 22 INDUSTRY Structural Analysis 23 Computational Fluid Dynamics 23 #####################________________________________________________________ Appendix V 25 THE CHEMICAL AND Molecular Modeling 26 PHARMACEUTICAL Structural Analysis 27 INDUSTRIES Computational Fluid Dynamics 27 #####################________________________________________________________ Appendix VI 28 COMPANIES INTERVIEWED REGARDING SUPERCOMPUTER USE #####################________________________________________________________ Appendix VII 29 COMPANIES INTERVIEWED REGARDING HIGH-SPEED NETWORK USE #####################________________________________________________________ Appendix VIII 30 MAJOR CONTRIBUTORS TO THIS REPORT ______________________________________________________ Abbreviations ARCO Atlantic Richfield Company GAO General Accounting Office GM General Motors Corporation IBM International Business Machines Corporation IMTEC Information Management and Technology Division NASA National Aeronautics and Space Administration NSF National Science Foundation Page 10 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks Appendix I _____________________________________________________________________________ OBJECTIVES, SCOPE, AND METHODOLOGY _____________________________________________________________________________ At the request of the Senate Subcommittee on Science, Technology, and Space, the Senate Committee on Commerce, Science, and Transportation; the House Subcommittee on Technology and Competitiveness; and the House Committee on Science, Space, and Technology; we reviewed various industries' use of supercomputers and high-speed networks. The purpose of our review was to (1) illustrate how the automobile, aerospace, petroleum, and chemical and pharmaceutical industries are using supercomputers to improve products, reduce costs, save time, or provide other benefits; (2) describe barriers that inhibit the increased use of supercomputers; and (3) provide examples of how certain industries use high-speed networks and their associated benefits. To illustrate how industries are using and benefitting from supercomputers and identify barriers to their increased use, we interviewed managers, scientists, and engineers from the 24 companies listed in appendix VI. We selected these companies on the basis of recommendations from various experts knowledgeable about industrial supercomputer use. Most of the companies we selected are Fortune 500 companies, largely because of the resources required to purchase, maintain, and use supercomputers. We also interviewed and obtained background information on supercomputers and on industry applications and future trends from industry analysts and consultants, hardware vendors, and government officials. The industry analysts and consultants included those from Research Consortium, Inc., Dataquest, The Superperformance Computing Service, Gartner Group, Inc., and the Institute for Supercomputing Research Recruit Co., Ltd. The hardware vendors included Cray Research, Inc., International Business Machines Corporation, Thinking Machines Corporation, and Silicon Graphics, Inc. The government officials included those from the Office of Science and Technology Policy, International Trade Administration, Department of Commerce, Lawrence Livermore National Laboratory, Department of Energy, and National Aeronautics and Space Administration Ames Research Center, National Science Foundation (NSF), and NSF supercomputer centers--San Diego Supercomputer Center, National Center for Supercomputing Applications at the University of Illinois at Urbana- Champaign, Cornell Theory Center, and Pittsburgh Page 11 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks _____________________________________________________________________________ Appendix I Objectives, Scope, and Methodology _____________________________________________________________________________ Supercomputing Center. We also interviewed and obtained documents from representatives of the Institute of Electrical and Electronics Engineers, Inc., and the American Petroleum Institute. To assess how industries use high-speed computer networks, we collected information from companies in various industries and procured the services of Texas Internet Consulting, a firm that designs and implements networks for large corporations. This firm subsequently interviewed the ten computer, automobile, and oil companies listed in appendix VII to determine how these companies use and benefit from high-speed networks. These companies had been selected based on the recommendation of experts as being frequent users of high-speed networks. Texas Internet Consulting also provided background information on high-speed networks and related applications. We discussed the information in this report with scientists, engineers, and other experts from 14 oil, aerospace, automobile, chemical and pharmaceutical, and computer companies and have incorporated their views as appropriate. However, we did not verify the validity or accuracy of the examples of dollar savings and productivity improvements provided to us by the various companies. In some cases, we were unable to obtain such information because it was considered proprietary and could not be released. In other cases, company representatives said they had not performed the extensive analysis necessary to quantify such benefits. Our review was conducted from October 1990 to May 1991 primarily in Washington, D.C., and other locations listed in appendix VI. Page 12 GAO/IMTEC-91-58 Supercomputers and High-Speed Networks Appendix II _____________________________________________________________________________ THE OIL INDUSTRY _____________________________________________________________________________ Many large oil companies worldwide--including American, British, French, and Middle Eastern companies--use supercomputers to better determine the location of oil and gas reservoirs, and to maximize the output from these reservoirs. The oil industry is among the early users of supercomputers, with ARCO being the first company to purchase a Cray supercomputer, in 1980, to model the largest oil reservoir in the U.S.--Prudhoe Bay, Alaska. Since that time, the oil industry has invested hundreds of millions of dollars in developing software for supercomputer applications. In addition, the industry uses off-the-shelf, consultant-developed and university-developed software. The oil industry is now looking at massively parallel processing to achieve the next-order-of-magnitude improvement in speed. The industry has many large computational problems--such as modeling entire reservoirs in greater detail--that cannot be practically attempted on today's supercomputers. Many companies are experimenting with parallel supercomputers and are converting vector software to para