Page 53

3
Recommendations

The committee believes that strong public support for a broadly based research program in information technology is vital to maintaining U.S. leadership in information technology. Facilitated access for both academic and industrial users to advanced computing and communications technologies has produced further benefits both in scientific progress and in U.S. industrial competitiveness. The committee's recommendations for the High Performance Computing and Communications Initiative (HPCCI) are based on this view of the importance of information technology to the country, as well as on the track record of success for the government's investment in information technology research. The committee's 13 recommendations address five different areas:

· General research program;
· High-performance computing;
· Networking and information infrastructure, including work focusing on the National Challenges;
· The supercomputer centers and the Grand Challenge projects; and
· Coordination and program management.

Within each area the recommendations are presented in priority order.

GENERAL RECOMMENDATIONS

As discussed in Chapter 1, government investment has played a major role in maintaining U.S. leadership in information technology and in helping to advance the technology, providing benefits to virtually every citizen. The return on federal investment has been substantial.

Recommendation 1. Continue to support research in information technology. Ensure that the major funding agencies, especially the National Science Foundation and the Advanced Research Projects Agency, have strong programs for computing research that are independent of any special initiatives. Past investment has yielded significant returns, as demonstrated in Chapter 1. Continued government investment in computing research, at least as high as the current dollar level, is critical to continuing the innovation essential to maintaining U.S. leadership.

Today the HPCCI supports nearly all of this research, an arrangement that is both misleading and dangerous: misleading because much important computing research addresses areas other than high performance (even though it may legitimately fit under the new Information Infrastructure Technology and Applications (IITA) component of the HPCCI), and dangerous



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



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 53
Page 53 3 Recommendations The committee believes that strong public support for a broadly based research program in information technology is vital to maintaining U.S. leadership in information technology. Facilitated access for both academic and industrial users to advanced computing and communications technologies has produced further benefits both in scientific progress and in U.S. industrial competitiveness. The committee's recommendations for the High Performance Computing and Communications Initiative (HPCCI) are based on this view of the importance of information technology to the country, as well as on the track record of success for the government's investment in information technology research. The committee's 13 recommendations address five different areas: · General research program; · High-performance computing; · Networking and information infrastructure, including work focusing on the National Challenges; · The supercomputer centers and the Grand Challenge projects; and · Coordination and program management. Within each area the recommendations are presented in priority order. GENERAL RECOMMENDATIONS As discussed in Chapter 1, government investment has played a major role in maintaining U.S. leadership in information technology and in helping to advance the technology, providing benefits to virtually every citizen. The return on federal investment has been substantial. Recommendation 1. Continue to support research in information technology. Ensure that the major funding agencies, especially the National Science Foundation and the Advanced Research Projects Agency, have strong programs for computing research that are independent of any special initiatives. Past investment has yielded significant returns, as demonstrated in Chapter 1. Continued government investment in computing research, at least as high as the current dollar level, is critical to continuing the innovation essential to maintaining U.S. leadership. Today the HPCCI supports nearly all of this research, an arrangement that is both misleading and dangerous: misleading because much important computing research addresses areas other than high performance (even though it may legitimately fit under the new Information Infrastructure Technology and Applications (IITA) component of the HPCCI), and dangerous

OCR for page 53
Page 54 because reduced funding for the HPCCI could cripple all of computing research. The "war on cancer" did not support all of biomedical research, and neither should the HPCCI or any future initiative on the nation's information infrastructure subsume all of computing research. Recommendation 2. Continue the HPCCI, maintaining today's increased emphasis on the research challenges posed by the nation's evolving information infrastructure. In addition to the work on infrastructure carried out in the new IITA program, continuing progress is needed in areas addressed by the HPCCI's other four components (High-Performance Computing Systems, National Research and Education Network (NREN), Advanced Software Technology and Algorithms, and Basic Research and Human Resources). The NSFNET and the gigabit testbeds have demonstrated the ability to build larger-scale, higher-performance networks, but ongoing research in several areas is still needed before a ubiquitous high-performance information infrastructure can be developed and deployed nationwide. The committee supports the HPCCI's increasing focus on information infrastructure, emphasizing that successful evolution of the nation's communications capability rests on continued investment in basic hardware, networking, and software technologies research. To further this evolution, which is consistent with administration efforts, including the addition of the IITA program, plus General Accounting Office (GAO, 1994) and other recommendations, the committee has identified in Recommendations 3 through 10 program areas that should receive (a) increased emphasis, (b) stay at present levels, and (c) have reduced federal support. RECOMMENDATIONS ON HIGH-PERFORMANCE COMPUTING Recommendation 3. Continue funding a strong experimental research program in software and algorithms for parallel machines. It is widely recognized that software for parallel computers lags behind hardware development. Progress in software and algorithms for parallel computers will determine how quickly and how easily we can use them.' A shift in emphasis toward increased funding for software and algorithm activities under high-performance computing has already begun. This shift properly reflects the urgency of investing more in software. The committee recommends the following approach to continue progress in research areas critical to developing and building needed software and algorithms: · Continue research on compilers, programming languages, and tools aimed at making it easier to use parallel computing machines. Critical needs include improved portability across machines, improved ability to run programs on machines of different sizes, and better understanding of how best to use different multiprocessor memory organizations. · Continue to develop experimental operating systems for parallel computers. More operating system experience will help us learn how to improve parallel hardware. Focus on the underlying research challenges posed by parallel machines rather than developing commercial operating systems technology. · Continue research on database and information systems for parallel machines. Such applications have increased in importance and represent a promising area for using parallel computing. · Continue research in the use of parallel computing for graphics and visualization. Graphics applications are valuable both because they demand much from their software and hardware and because they stimulate effective use of high-performance computing by

OCR for page 53
Page 55 offering computational scientists and other end-users the ability to analyze complex data and problems. · Fund sufficient hardware purchases to ensure needed access for computer scientists and end users trying to evaluate the effectiveness of new architectures and software technologies. Dedicated access to expensive machines is often required for operating systems development or for controlled measurement of software performance, and sometimes dedicated access is needed to full-scale machines, which are then most economically housed in a centralized, national center. The importance of the local availability of mid-range machines for researchers in software for parallel computers was noted in the Branscomb report (NSF, 1993). · Provide resources to help complete the development and distribution of compilers, programming tools, and related infrastructure broadly usable by the software research community. Such infrastructure—which may be developed by individual research groups or by centers (such as the NSF science and technology centers)—has been crucial to rapid progress. For example, tools for the design of very large scale integrated (VLSI) circuits allowed many researchers to undertake VLSI designs. The committee notes that funding agencies should avoid turning related infrastructure development efforts into product development efforts. · Seek improved integration of parallel computing hardware and software with communication networks, both in software and hardware research. · Emphasize design and analysis of new algorithms for parallel computing, as well as implementation and evaluation of these algorithms on real parallel machines. Opportunities for development of new parallel algorithms exist in both scientific and information infrastructure-related applications. The theoretical performance and scaling efficiency of new algorithms need to be demonstrated by actual implementation and evaluation on parallel machines, first by computer scientists and then embedded in real end-user applications. · Ensure that effective new algorithms for parallel computing are made widely available to end-user communities to assist in building applications. Recommendation 3.1. Avoid funding the transfer ("porting") of existing commercial applications to new parallel computing machines unless there is a specific research need. Several existing applications enjoy widespread commercial use on large uniprocessor and vector machines; examples include third-party codes in chemistry, biomolecules, engineering fluid dynamics, deformable structures, and database access. It has been proposed by some that the HPCCI should fund transferring, or "porting," such applications to new types of parallel computers as a way to enhance the attractiveness of new parallel machines. The committee finds inappropriate the use of federal HPCCI funding for such porting of applications for several reasons. First, the algorithms used in these applications were designed for sequential or vector computing, and thus little new knowledge will be gained from merely porting existing applications to a parallel machine without redesigning the algorithms. Second, the open market will fund such transfers if a sufficient user base exists. Third, choosing whose application to transfer and to which machines will involve the HPCCI in picking winners from among many commercial vendors. Although it recognizes that a federal agency might decide that one of its missions would best be served by porting an existing application to a parallel computer, the committee recommends that funding of such ports be justified on the basis of the agency mission and not as HPCCI

OCR for page 53
Page 56 research. The committee believes that it is legitimate for groups of agencies to work together to develop community codes for common applications needed by their several missions. Likewise, carrying out an HPCCI research program may require that applications be available on a particular parallel machine, in which case the transfer could be justified by the importance of the research it enabled. Finally, the committee also sees a legitimate reason to port existing applications for the purpose of evaluating machines within a research laboratory or university. Recommendation 4. Stop direct HPCCI funding for development of commercial hardware by computer vendors and for ''industrial stimulus" purchases of hardware. Maintain HPCCI support for precompetitive research in computer architecture; this work should be done in universities or in university-industry collaborations and should be driven by the needs of system and application software. The development and placement of parallel hardware to date was necessary to establish parallel computing as a viable alternative to sequential and vector computing. (The establishment of this paradigm is discussed throughout Chapter 2 and in Appendixes A and E.) Industry is now willing and able to improve on the base of ideas established by the HPCCI, at least for mainstream parallel machines (special government requirements are discussed below). Government development funds should no longer be spent in industry either to further refine parallel machines or to purchase machines as a stimulus for vendors. The committee notes that use of HPCCI funds for these purposes has already decreased significantly, a trend that the committee supports.2Federal funding of hardware developments within companies should continue to decline, unless some special agency need demands the development of nonstandard hardware (e.g., a high-performance system for use on a ship or in an airplane, such as Intel Corporation's parallel Paragon computer made more rugged for military use, or for a highly specialized application). In such cases, agency mission funds, and not HPCCI funds, should be used. Important precompetitive hardware research problems merit continued federal funding because the development of parallel computing architecture and gigabit networks will not be the final chapter in the continuing development of ever more powerful systems. The committee recommends that ongoing research efforts in hardware and architecture be based in academic and research institutions, possibly in collaboration with industry. Potential problems can be minimized if the research institution serves as the project lead, and if the research challenges rather than commercial development are the focus (Cohen and Noll, 1994). Not only do academic institutions have more freedom to think about longer-term issues, but they also stimulate technology transfer through publication and placement of graduates. The national experience supports a basic tenet of Vannevar Bush: publicly funded research carried out in universities produces excellence, diversity, fresh ideas, trained people, and technology transfer (OSRD and Bush, 1945). Commercial organizations, on the other hand, have powerful incentives to avoid distributing new ideas widely and may even impede the introduction of new technology when it competes with existing products. To narrow the gap between parallel computing hardware capabilities and the software needed to use them, research on architecture should be driven by software and applications needs. Thus, further integration of application and system software needs into architecture research should be encouraged in any funding of architecture research. Recommendation 5. Treat development of a teraflop computer as a research direction rather than a destination. The committee believes that federal investment in developing or purchasing machines to demonstrate raw scalability for its own sake is inappropriate, except as a focus for precompetitive, academic research. Instead, the focus should be on matching agencies' mission requirements to the emerging sustainable scalable architectures. Such architectures will very likely reach 1-teraflop capability before the end of this decade using 1,000 or so high-performance commercial microprocessors. The goals of scalability over many sizes of machine and of demonstrating teraflop performance have been useful in pointing toward the use of mass-produced devices in large

OCR for page 53
Page 57 collections to solve complex computing tasks, but implementation of a machine of any specific size can be premature. Moreover, seeking a common design over a large size range is wasteful because the expensive communication paths required to harness large numbers of inexpensive processors together are inappropriate when scaled down to smaller machines with only a few processors. The pursuit of wide scalability may have deferred early consideration of shared-memory parallel computers, the type that today appears promising. In fact, the focus on teraflop capability detracts from other important aspects of high performance, such as memory and input/output systems, which are critical components of any high-performance system. Advances in parallel architectures together with progress in the underlying integrated circuit technology will continue to provide improvements in performance/cost ratios that will naturally bring computing power to the teraflop level. Most industry analysts see the potential for single microprocessors with 1- to 2-gigaflop peak performance by the end of the decade. Combining 512 to 1024 such future microprocessors in a scalable system would create a teraflop capacity at roughly the price of today's supercomputers, with capabilities of tens of gigaflops possible. Supporting research into the key technologies needed to achieve and use scalable computing, combined with patience to see how the relative economics of computing power and communications interact, seems to this committee to be the most efficient approach to increasing performance. The committee thus emphasizes that the HPCCI should treat the goal of teraflop performance as a milestone to be reached naturally by computer vendors in due course, not on a forced time scale. The HPCCI should continue to fund research on technologies that will contribute to reaching the goal. At some point in the near future a teraflop parallel machine will be built when some agencies' mission requirements correspond to a sufficiently economical commercial offering. Continued progress will naturally lead to machines much larger than a teraflop. RECOMMENDATIONS ON NETWORKING AND INFORMATION INFRASTRUCTURE The committee believes that the successes of the HPCCI in establishing scalable compute servers, investigating high-performance networks, and forming interdisciplinary teams of computer and application scientists are setting the stage for important new research to support enhancement of the nation's information infrastructure. An increased emphasis on the research needed to achieve such an infrastructure is desirable (CSTB, 1994d). In fact, this shift has already begun; spending on networking and IITA activities accounted for nearly 50 percent of the HPCCI budget requested in FY 1995:3 $177 million for the NREN and $282 million for IITA. This is a significant increase over the $114 million that was spent for NREN in FY 1993, the year prior to the addition of IITA. The committee believes that such a shift is appropriate. Recommendation 6. Increase the HPCCI focus on communications and networking research, especially on the challenges inherent in scale and physical distribution. Advancing the nation's information infrastructure will put great demands on digital communications technology for providing broad access to services. Ensuring broad access poses a host of technical and economic questions for which existing solutions are inadequate. The committee recommends increased support for learning how to attach millions of users to a digital communications structure that provides a wide array of services and greater integration of services, and how to accommodate the demands that these users will generate using the novel applications enabled by such an information infrastructure. Recommendation 7. Develop a research program to address the research challenges underlying our ability to build very large, reliable, high-performance, distributed information systems based on the existing HPCCI foundation. An improved infrastructure will need to offer

OCR for page 53
Page 58 capability to all facets of our economy on a scale not yet imagined, and no one yet anticipates all of the ways that users will use such an information infrastructure. Improvements to the nation's information infrastructure and activities related to it have generated a level of public interest matched by only a few technology-based objectives. The committee is concerned that unrealistic expectations for availability and for the quality and range of services could encourage a short-term, product-oriented focus in funding research activities4 that would not be in our country's best interest. Care should be taken to apprise policymakers and the public of the long time needed for development and wide-scale deployment of the services expected to be available through the information infrastructure. The committee strongly recommends that the HPCCI remain focused on the basic research issues arising from desired improvements to the information infrastructure, evolving from its early emphasis on parallel, high-performance computing, high-performance networking, and scientific applications to the broader issues of connection, scale, distributed systems, and applications. The addition of the IITA area to the HPCCI was a key step in accelerating a shift in focus of the research community to the challenges of improving the nation's information infrastructure. The committee has identified three key areas where new emphasis is critical to supporting the research needs associated with the information infrastructure: 1. Scalability. While the HPCCI has emphasized large computing systems on the order of thousands of interacting computers, an enhanced, nationwide information infrastructure will require scaling to millions of users. In addition, the HPCCI has emphasized bringing the highest performance to bear on individual scientific applications, whereas improving the information infrastructure for the nation requires using the highest performance to meet the practical needs of millions of simultaneous users. 2. Physical distribution and the problems it raises. A better information infrastructure will emphasize geographical distribution with its limitations on bandwidth, increase in latency of communication, and additional challenges in secure and reliable communication. These challenges have been much less severe in localized high-performance parallel systems. Research on both distributed and parallel systems technology will be important in supporting this aspect of a national-scale information infrastructure. 3. Innovative applications. A shift should occur from a focus on specific Grand Challenge problems in science to well-formulated National Challenges that affect a wider segment of society. The committee sees an important role for development and demonstration of easily appreciated applications that will drive the technology of the information infrastructure. Improving scalability and physical distribution requires investment in both: · Hardware and architecture, including systems that efficiently handle a rich mix of text, images, and audio and video data; systems that provide cost-effective, high-bandwidth, end-to-end communications; and systems that provide information access to large numbers of users; and · Software, including basic networking software for encryption, routing, flow control, and so on; tools for providing and building such capabilities as scheduling, bandwidth optimization, video handling, and service adaptation; and many others. This is the so-called "middleware."

OCR for page 53
Page 59 The committee believes that building on the current HPCCI model of a coordinated program, avoiding central control, seems even more crucial for the IITA portion of the research program, because the challenges posed by an information infrastructure are inherently less well defined than those addressed by other components of the initiative. The committee is encouraged by the early development of cooperative research programs in IITA areas, such as the digital libraries program, which includes NSF, ARPA, and the National Aeronautics and Space Administration (NASA), and by recent attempts to identify topics for research, such as discussions among several hundred researchers and others at a workshop in early 1994 (Vernon et al., 1994). Recommendation 8. Ensure that research programs focusing on the National Challenges contribute to the development of information infrastructure technologies as well as to the development of new applications and paradigms. The National Challenges incorporate socially significant problems of national importance that can also drive the development of information infrastructure. Hardware and software researchers should play a major role in these projects to facilitate progress and to improve the communication with researchers developing basic technologies for the information infrastructure. Awards to address the National Challenges should reflect the importance of the area as well as the research team's strength in both the applications and the underlying technologies. The dual emphasis recommended by the committee contrasts with the narrower focus on scientific results that has driven many of the Grand Challenge projects. Because the National Challenges as currently defined are too broad and vague to offer specific targets for large-scale research, the notion of establishing testbeds for a complete national challenge is premature. Instead, research funding agencies should regard the National Challenges as general areas from which to select specific projects for limited-scale testbeds or focused software research projects. Particular areas in which a focused research target can be identified (e.g., the ARPA-NSF-NASA digital library testbeds) may be appropriate for slightly higher funding, but the committee believes that very large scale applications development is premature. At this early stage, letting "a thousand flowers bloom" will provide a better pay-back than funding a few large or fullscale deployments. (Box A.3 and related text in Appendix A give more information on the National Challenges.) RECOMMENDATIONS ON THE SUPERCOMPUTER CENTERS AND GRAND CHALLENGE PROGRAM The four NSF supercomputer centers are the largest single element of the FY 1995 HPCCI implementation plan in dollars ($76 million, or 6.6 percent of the requested FY 1995 HPCCI budget) and the largest infrastructure project in the initiative. The centers—which give users access to a broad array of powerful tools ranging from highly innovative to mature—are a significant national resource for gaining knowledge, experience, and capability. Thanks to their leadership, high-performance computing and communications are now widely accepted as an important tool in academia, industry, and commerce. The centers have played a major role in establishing parallel computing as a full partner with the prior paradigms of scalar and vector computing. They have contributed by providing access to state-of-the-art computing facilities to a broad range of users. As new large-scale architectures appear, the centers stimulate their early use by providing access to these architectures and by educating and training users. (Appendix E details the accomplishments of the NSF centers and of their national user base.) The committee recognizes that advanced computation is an important tool for scientists and engineers and that support for adequate computer access must be a part of the NSF research program in all disciplines. The committee also sees value in providing large-scale, centralized computing, storage, and visualization resources that can provide unique capabilities. How such

OCR for page 53
Page 60 resources should be funded and what the long-term role of the centers should be with respect to both new and maturing computing architectures are critical questions that NSF should reexamine in detail, perhaps via the newly announced Ad Hoc Task Force on the Future of the NSF Supercomputer Centers Program. For example, much of the general access to computing resources at the centers is provided on maturing architectures. Neither the quality of the science performed by the users of such technology nor the appropriateness of NSF funding for such general access is in question. The committee did not consider the appropriate overall funding level for the centers. Nonetheless, the committee does question the exclusive use by the NSF of HPCCI-specific funds for support of general computing access, which in itself does not simultaneously help drive the development of high-performance computing and communications technology. In this regard, NSF follows a different approach to funding its supercomputer centers than do NASA and the Department of Energy (DOE), where HPCCI funds are used only to support the exploration and use of new computing architectures, while non-HPCCI funds are used to support general access. The committee believes that DOE's and NASA's approach to funding general access should be followed across the agencies. Also, as the committee points out in Recommendation 12, including all of the NSF supercomputer centers' funding under HPCCI could cause major disruption to the centers' national mission should the HPCCI be altered significantly. Recommendation 9. The mission of the National Science Foundation supercomputer centers remains important, but the NSF should continue to evaluate new directions, alternative funding mechanisms, new administrative structures, and the overall program level of the centers. NSF could continue funding of the centers at the current level or alter that level, but it should continue using HPCCI funds to support applications that contribute to the evolution of the underlying computing and communications technologies, while support for general access by application scientists to maturing architectures should come increasingly from non-HPCCI funds. Examination of the supercomputer centers program should include identification of: · Emerging new roles for the centers in supporting changing national needs; and · Future funding mechanisms, including charging mechanisms and funding coupled to disciplinary directorates. In addition to enabling high-performance scientific computing, several of the NSF centers have developed new software technologies that have significantly affected other parts of the HPCCI. The most obvious of these is the recently developed Mosaic World Wide Web browser. The committee recommends that NSF continue to take a broad view of the centers' mission of providing access to HPCCI resources, including, but clearly not limited to, participating in research needed for improved software for parallel machines and for enhancement of the nation's information infrastructure. The centers, and the researchers who use their facilities, should compete for research funds by the normal means established by the funding agencies. Recommendation 10. The Grand Challenge program is an innovative approach to creating interdisciplinary and multi-institutional scientific research teams; however, continued use of HPCCI funds is appropriate only when the research contributes significantly to the development of new high-performance computing and communications hardware or software. Grand Challenge projects funded under the HPCCI should be evaluated on the basis of their contributions both to high-performance computing and communications and to the application area. The Grand Challenge problems are sufficiently large and complex and the research teams addressing them are capable enough to exercise the parallel computing technology thoroughly and to test its capability. These efforts have been supported under the HPCCI as a valid way to involve real users in parallel computing, but as parallel computing becomes an established tool, the need to use the HPCCI to stimulate the user community will decrease. Furthermore, the use of

OCR for page 53
Page 61 high-performance computing will become more pervasive, making it unreasonable to include all such programs with the HPCCI. The committee recommends completion of the initial Grand Challenges as planned over the next few years. Currently, although the scientific disciplines are providing major funding for Grand Challenge projects (e.g., more than 50 percent of requested FY 1995 funds for NSF Grand Challenges come from the scientific and engineering research directorates), virtually all of the Grand Challenge funding is labeled HPCCI. The committee urges that any follow-on funding of Grand Challenges include a significant and growing fraction of non-HPCCI scientific disciplinary funds. This will limit the selection to tasks whose scientific interest justifies their cost, in competition with other research in their respective disciplines. The committee sees an ongoing value from the strong interaction between challenging applications and new architectures and software systems and from cooperation between computer and computational scientists—a number of the Grand Challenge teams have demonstrated that collaboration can lead to advances in both computing and the particular scientific discipline involved. Partial funding of applications research that contributes to the development of new hardware and software systems is a legitimate use of HPCCI funds. Such activities must be evaluated on the basis of their contributions both to high-performance computing and communications technologies and to the application area. RECOMMENDATIONS ON COORDINATION AND PROGRAM MANAGEMENT IN THE HPCCI Recommendation 11. Strengthen the HPCCI National Coordination Office (NCO) while retaining the cooperative structure of the HPCCI and increasing the opportunity for external input. As the committee pointed out in its interim report (CSTB, 1994c, p. 9), the dimensions of the need for clear communication about the HPCCI have recently become apparent: congressional oversight activities and other indicators suggest that the HPCCI is of concern to a growing constituency and that often a variety of audiences need detailed explanations of it. Such an effort will add substantially to the work of the NCO, which has been headed by a half-time, permanent-position director who holds a concurrent, half-time appointment as director of the National Library of Medicine (NLM).5The other NCO staff positions are a mix of permanent positions, contract positions, and temporary positions filled by individuals on loan from other federal agencies for limited periods of time, often no more than 1 year.6 Although the NCO reports to the Office of Science and Technology Policy (OSTP) on programmatic matters, administrative functions such as office space, salaries, and benefits have been handled through the National Institutes of Health. The temporary nature of some of the NCO positions jeopardizes continuity and cumulative insight. Further, limited staff resources raise questions about the NCO's capacity to meet the challenge of the growing volume, complexity, and urgency of the outreach efforts needed for the initiative (CSTB, 1994c, p. 9). The NCO serves an important coordination and communication function both among agencies of the government and between the agencies, Congress, industry, and the public. It is to the credit of the NCO staff that the HPCCI has been an effective model of interagency collaboration. In recommending a strengthening of the NCO, the committee strongly endorses the current NCO's role of supporting the mission agencies rather than directing them. The committee believes that it is vital that direction of the HPCCI remain in the agencies. By avoiding actual direction the NCO leaves mission judgments in the hands of responsible agency officials who are accountable for the allocation of their resources. By avoiding the appearance of direction the NCO encourages an appropriate diversity of research projects as each agency capitalizes on its best ideas. Mission agencies cooperate effectively with each other and

OCR for page 53
Page 62 with the NCO precisely because it does not threaten their autonomy. This cooperation could easily vanish were the NCO seen as functioning with too heavy a hand. The committee believes that the value of interagency cooperation outweighs the potential benefits that might be gained through more centralized management of the HPCCI (CSTB, 1994c, p. 8). The committee strongly recommends retaining the model of a cooperative and coordinated interagency program. Some individuals and organizations have expressed concern about the lack of centralized management of the HPCCI. However, the committee believes that the current cooperative structure is one of the initiative's strengths, providing room for diversity of thought and action. Such diversity is essential to progress, because no one manager or agency has a monopoly on the right ideas for the future of computing and communications. Central management of the HPCCI could focus its activities too narrowly, as well as lead to reduced interest in the program by agencies that found that the centralized agenda no longer matched their interests.7 The committee believes that government investment in information technology research has often enjoyed first-class leadership. Program officers with vision have supported innovative ideas, leading to later successes. The committee emphasizes that the best method for making continued research investment is to continue to attract highly competent program officers and to give them the flexibility to develop effective programs. In the past, this approach has yielded solid returns on the research investment. Furthermore, it has encouraged necessary diversity in the research program, thus increasing opportunities for unexpected discoveries and ensuring a broad perspective in addressing problems. Recommendation 11.1. Immediately appoint the congressionally mandated advisory committee intended to provide broad-based, active input to the HPCCI, or provide an effective alternative. The HPCCI could be improved by input from and review by an advisory committee with balanced representation from industry and academia, including current and potential users and developers of high-performance computing and communications. If appointment of such a committee is not feasible, some alternate mechanism should soon be developed to provide similar input. The committee is aware of the recent increases in the number of advisory committees, as well as the danger of having too many committees. Thus, the committee recommends that the HPCCI advisory committee have a well-defined role focusing primarily on providing external input into the coordination and planning for the HPCCI. Recommendation 11.2. Appoint an individual to be a full-time coordinator, program spokesperson, and advocate for the HPCCI. Having a part-time NCO director has served well to this point, but the broadening of the HPCCI demands leadership unencumbered by other major responsibilities. A full-time person could either direct the NCO or could report to the director and would work to strengthen the ties between the HPCCI, industry, the Congress, and the public. The committee uses the word "coordinator" rather than "manager" to emphasize the need for coordination and communication that avoid usurping the authority of the mission agencies. The NCO should remain within the OSTP structure. Recommendation 12. Place projects in the HPCCI only if they match well to its objectives. A number of preexisting agency programs have entered the HPCCI. Such administrative changes make the HPCCI budget appear to grow faster than the real growth of investment in high-performance computing and communications. Some of these programs exactly match the goals of the HPCCI and are properly included. Others are only marginally relevant and might better be placed elsewhere in agency budgets. The committee sees the possibility of a longterm danger to important programs, such as basic research in computing within NSF and ARPA, should the HPCCI end. Recommendation 12.1. Federal research funding agencies should promptly document the extent to which HPCCI funding is supporting important long-term research areas whose future funding should be independent of the future of the HPCCI. The committee found that many research areas predating the HPCCI and related only partly to its goals are now under the

OCR for page 53
Page 63 HPCCI umbrella. Although encouraging important research areas to include and even focus on HPCCI-related components, the process of coding all funding in a research area as high-performance computing and communications can be dangerous. In many cases, areas were recoded as high-performance computing and communications without receiving an increase in funds. The danger in this process is that changes in the direction or level of funding for the HPCCI could lead to unintentional changes in the funding levels of important research areas, even if they are largely unrelated to the HPCCI and even if they have received none of the HPCCI incremental funding! This problem is particularly acute at NSF, where nearly all of the funding in the Computer and Information Science and Engineering directorate is coded as HPCCI funding. Given that NSF is not a mission agency and is investigator-driven, this approach seems shortsighted. NSF would have to retain funding for computer science research even in the absence of the HPCCI. Ongoing funding of important research areas in computer science will be critical to the nation's future, independent of the future of the HPCCI. Recommendation 13. Base mission agency computer procurements on mission needs only, and encourage making equipment procurement decisions at the lowest practical management level. To stimulate the use of parallel computing early in the HPCCI's 5-year time frame, it has been appropriate and necessary to place into service a reasonable number of highly parallel machines for serious algorithm and software development. Early development of an adequate base of parallel computers was essential to shifting the attention of industry and research organizations toward parallel computing. Now, however, it is more appropriate to base procurement of computer hardware on mission needs only. One program that claims to have done so already is the Defense High Performance Computing Modernization Program. The committee applauds modernization of the computing facilities available to Department of Defense organizations and the mission-driven nature of the procurement process, which should be established in all agencies. Each agency has responsibility for its own budget and its own requirements. The committee believes that agencies should base procurements of computing equipment on their needs and budget constraints. Agencies should be free to purchase parallel computers when they suit agency needs. Individual agencies can balance the cost of obtaining applications against the cost of computing equipment so as to best match procurements to their requirements. Parallel computing is now mature enough to be considered a viable alternative to other forms of computing and may deliver suitable computing power at less cost than other architectures. Although the committee firmly recommends that computer purchases be guided by mission needs, it also sees a role for collaboration between mission agencies and industrial or university parallel computing consortia. Direct agency responsibility for missions, budgets, and equipment purchases can be reconciled with the advantages of group action through participation in appropriate consortia. For example, the NSF centers have proven valuable for offering exploratory experience with high-performance computing, and it is encouraging to see industrial-academic consortia forming to explore parallel computing. The committee encourages mission agencies to participate with the NSF centers and other parallel computing consortia. Such participation offers knowledge at low cost and leads ultimately to more cost-effective procurements. The committee's recommendation that equipment be selected at the lowest practical management level applies equally to government agencies and to government contractors. The direct manager of a computing facility is charged with making it work and will do that task best if allowed to select equipment that matches the facility's needs. The committee believes that agencywide procurement of standard brands, while promoting collaboration, can weaken the responsibility of the user organizations. Likewise, it has generally been best for an agency to specify the results it wants and to leave the choice of specific equipment to the contractor.8 Delegating equipment selection not only saves top-level agency decision-making resources but also places responsibility for purchase decisions firmly in the hands of the managers who must deliver results.

OCR for page 53
Page 64 COMMENTS RELATING THIS REPORT'S RECOMMENDATIONS FOR HIGH-PERFORMANCE COMPUTING AND COMMUNICATIONS RESEARCH TO ADMINISTRATION PRIORITIES This report's recommendations broadly address much of the computer science and engineering research being conducted today, as well as the HPCCI specifically. In a May 1994 memorandum from the director of the Office of Management and Budget to all agency heads, the administration outlined its priorities for U.S. research and development in general. Box 3.1 briefly compares applicable parts of that memorandum to the positions taken and actions recommended in this report of the Committee to Study High Performance Computing and Communications. BOX 3.1 Comparison of Administration Priorities for Harnessing Information Technology to Committee Recommendations in This Report Computing Systems "The development of scalable systems with the input/output capabilities, mass storage systems, real-time services, and information security features needed to build and fully utilize the National Information Infrastructure (NII) should be emphasized. High performance computing systems capable of 1012 operations per second (a teraflop) on technical problems will be achieved by FY 1997. Emphasis should be placed on advances in information storage media for both high and low end applications; systems integration of clustered workstations and large parallel systems; development of advanced tools and processes for the design and prototyping of faster semiconductor devices; and research on nanotechnology, photonics, flat panel displays, and integrated micro-electrical-mechanical devices." The Committee on High Performance Computing and Communications recommends continuing the focus on high-performance parallel computing, but decreasing the emphasis on achieving a teraflop computing system on a specific time scale. The committee agrees that scalable high-performance computing and communications technology that supports the nation's emerging information infrastructure is important. The committee also recommends increased emphasis on the support of communications within new computer systems. Networking and Communications "It will be necessary to support the development of the networking technology required for the deployment of national gigabit speed networks incorporating heterogeneous carriers including satellite and wireless capability. This means serving hundreds of millions of users and demonstrating mobile and wireless capability. It includes the development of interoperability concepts and technologies and the integration of computers, televisions, telephones, wireless telecommunications and satellites." The committee thoroughly agrees. continues

OCR for page 53
Page 65 BOX 3.1—continued Software, Algorithms, and Basic Research "The United States should conduct basic research to support the computational requirements of new computing paradigms. There is a need for new methods for data authentication and software verification and validation. The development of tools and techniques to enable assembly of systems from inexpensive, versatile, reusable software components is required." The committee's recommendations emphasize the importance of a complete program of basic research in computer science and underscore the importance of research on software and algorithms to make the best use of new computing paradigms. Information Infrastructure Services "Access to and utilization of the NII will require services, tools, and interfaces that facilitate a wide range of applications. These include registries, directories, navigation and resource discovery tools, data interchange formats, and other information services that help users find and query services and components in distributed repositories. There will have to be new types of human-computer access and the development of improved collaborative software, groupware, and authoring tools for multimedia will be important. Equally important are the development of privacy and security technologies and integrated information and monitoring systems." The committee enthusiastically agrees. Human-Computer Interaction ''New products and applications are enabled by those hardware and software technologies that will allow every American to use easily the NII. Development and use of the following should be advanced: virtual reality; simulation; flat-panel displays; video and high definition systems; three-dimensional sound, speech interfaces, and vision." The committee did not study human-computer interaction and specific related areas for research, although these are certainly a key part of a broad research program on information infrastructure. Computing and Communications Applications "The FY 1996 R&D budget should advance applications of high-performance computing and the NII. The Federal High-performance Computing and Communications Program is helping to develop the technologies and techniques needed to solve critical research problems that require more advanced computers, storage devices, algorithms, and software tools. Additional effort is needed to accelerate the transfer of these technologies from the laboratory to the marketplace." The committee recommends that applications be funded increasingly from outside the HPCC1, especially as the technology underlying the application becomes mature. HPCCI funds should focus research support on applications that affect the base computing and communication technologies, as well as solve new applications problems. The committee recommends that such a policy be followed in funding work on the Grand Challenges and the emerging National Challenges. SOURCE: Panetta (1994), pp. 10-11.

OCR for page 53
Page 66 NOTES 1. See the subsections "Programming" and "Algorithms" in Appendix A for a discussion of development and achievements to date. 2. Based on briefings to the committee by Anita Jones (Department of Defense); Duane Adams, Howard Frank, and John Toole (Advanced Research Projects Agency); and Victor Reis (Department of Energy). 3. See NCO (1994), p. 15. Note that figures represent the President's requested budget authority for FY 1995. Actual appropriated levels were not available at press time. Because the HPCCI is synthesized as a cross-cutting multiagency initiative, there is no direct "HPCCI appropriation." 4. This risk is illustrated in the GAO examples of standards setting and other nonresearch activities under the HPCCI umbrella. 5. A January 6, 1995, press release from the Office of Science and Technology Policy announced the resignation of the NCO's director. Donald A.B. Lindberg. Lindberg, requesting that a successor be named when his current 2-year term ends, recommended that a full-time director be appointed at this point in the evolution of the HPCCI. 6. Letter dated August 8, 1994, to Marjory Blumenthal (CSTB) from Donald A.B. Lindberg (NCO/NLM) in response to the committee's interim report (CSTB, 1994c). 7. The committee shares OSTP Director John Gibbons' concerns about the centralized management advocated by GAO (1994, p. 34). 8. The committee notes the extreme fruitfulness of this model in the hands of the late Sidney Fernbach of DOE's Lawrence Livermore National Laboratory, who for a generation kept his laboratory at the forefront of computing and at the same time helped to stimulate the development of generations of supercomputers.