Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
7 Institutional Infrastructure of Academic CS&E The term "institutional infrastructure" is used here to refer to the institutions that have some important bearing on academic CS&E. Thus institutional infrastructure includes major funding agencies that support research, the universities that house academic CS&E, and the various professional organizations that provide vehicles for dissemi- nation of research and other support to the discipline. FEDERAL AGENCIES FUNDING COMPUTER SCIENCE AND ENGINEERING An overview of federal support for CS&E was provided in Chap- ter 1. A more detailed description of each major research-supporting agency is provided below. (Figures cited are presented in constant 1992 dollars and are subject to the caveats specified in Note 18, Chapter 1.) Department of Defense The modern military is highly dependent on computers in almost every aspect of its responsibilities, including weapons acquisition, ~ and ~nntr~1 rommilnir~ti`~n~ intelligence weapons con ~VllLLlL~AL~ ~' t~ _~^~-~ ~ HA ~ ~ __w~ trot, and administration. Among federal agencies, the Department of Defense is the largest single funder of CS&E research; historically a little over one-third of 217
218 COMPUTING THE FUTURE this money has gone to universities and colleges, making the Depart- ment of Defense the largest supporter of academic CS&E research as measured by dollar amounts. Figure 7.1 illustrates the Defense De- partment's history of funding CS&E research for the last 15 years. Within the Department of Defense, the Defense Advanced Re- search Projects Agency (DARPA) is responsible for the majority of 500 400 In o 300 Cal o n o - ._ ~100 To all recipients 200 - o ~1 1977 1979 1981 f JO-en' lo. _ To academia /~~ / 1983 1985 1987 1989 1991 Fiscal Year FIGURE 7.1 Department of Defense obligations for research for computer science (basic and applied), FY 1976 to FY 1991, in constant FY 1992 dollars. SOURCE: Basic data (in then-year dollars) for all recipients taken from Fed- eral Funds for Research and Development (Federal Obligations for Research by Agency and Detailed Field of Science/Engineering: Fiscal Years 1969-1990), Divi- sion of Science Resource Studies, National Science Foundation. Data for FY 1990 taken from Federal Funds for Research and Development: FY 1989, 1990, and 1991, National Science Foundation, NSF 90-327. Data for FY 1991 are preliminary and were supplied to the committee by the Division of Science Resource Studies, National Science Foundation. Basic data (in then-year dol- lars) for academia taken from Federal Fundsfor Research and Development (Fed- eral Obligations for Research to Universities and Colleges by Agency and Detailed Field of Science/Engineering: Fiscal Years 1969-1990), Division of Science Re- source Studies, National Science Foundation. Figures include both "comput- er science" and "mathematics and computer science, not elsewhere classi- fied." Constant dollars calculated from GNP deflators used in National Science Foundation, Science and Engineering Indicators, 1991, NSF, Washington, D.C., 1991, Table 4-1.
INSTITUTIONAL INFRASTRUCTURE OF ACADEMIC CS&E 219 CS&E research. Other important roles are played by the science of- fices of the various services, the Office of the Secretary of Defense, and the National Security Agency. The influence of DARPA on CS&E has been pervasive. Founded in 1958 to promote research in fields of military interest, DARPA has been directly involved in supporting time-sharing (1960s), networks (late 1960s to mid-1980s), artificial intelligence (1970 to present), ad- vanced computer architectures and very-large-scale-integration cir- cuitry (1970 to present), and graphics (mid-1960s). In recent years, the major areas of CS&E concern to DARPA have included high-performance computing, networks, software, artificial intelligence (AI), and applications of these areas. DARPA divides its overall computing program into science (including machine transla- tion, scalable software libraries for high-performance computing, soft- ware understanding for the future), technology (including speech un- derstanding, knowledge representation, embedded microsystems), and applications (including image understanding, natural language pro- cessing, transportation planning). DARPA has long had a reputation for supporting high-risk, high- gain research in pursuit of military applications. Its style of research support is highly proactive in that DARPA identifies areas of poten- tial interest for military needs and orients its research support mostly toward experimental and prototype system development. Individual program managers have been highly influential, both in articulating areas of need and in stimulating the CS&E community to be interest- ed in these areas. Thus DARPA has often played a key role in defin- ing research agendas for the CS&E field. In the past, DARPA tended to concentrate its support in a few selected institutions, thereby creating an infrastructure of centers of excellence with critical masses of interested and active researchers. However, since the mid-1980s DARPA has been required to engage in competitive procurement practices, even for the award of con- tracts for basic research. This requirement has broadened somewhat the number of institutions receiving DARPA funding in CS&E but has also increased the administrative burdens (e.g., by insisting on more precise definition of deliverables than before) on established centers even though they may have demonstrated records of excel lence and success. Other agencies within the Department of Defense fill somewhat more specialized niches. For example, the Office of Naval Research (ONR), the Air Force Office of Scientific Research (AFOSR), and the Army Research Office (ARO) fund small but important research pro- grams in CS&E. In contrast to DARPA's emphasis on experimental
220 COMPUTING THE FUTURE and prototype work, these offices tend to emphasize relatively small- scale concept and algorithm development oriented toward the funda- mental science that will underlie future military applications. Rather than covering CS&E comprehensively, their research portfolios thus depend strongly on judgments about what these future applications will entail. The early ONR and the AFOSR had a tremendous impact on the development of computers in the 1940s and 1950s (Box 7.1~. Budgets for CS&E research within these offices are about 5 to 10 percent that of DARPA. The ONR research program includes activi- ties in software design and construction, distributed and parallel sys- tems, database systems, AI and robotics, real-time computing, fault tolerance, high-performance computing, and secure computing. In the near future, ONR expects to focus on dependable multicomputer systems, mathematical logics for programming languages, case-based reasoning, massively parallel computing for the physical sciences, algorithmic structural complexity, and visual processing. AFOSR's scientific program includes a variety of mathematical areas of inter
INSTITUTIONAL INFRASTRUCTURE OF ACADEMIC CS&E 221 est (e.g., dynamics, control theory, statistics, and signal processing) and fundamental computer science as well. The ARO supports work on high-performance computing, intelligent systems, artificial intelli- gence, and software. The Office of the Secretary of Defense (OSD) is the umbrella su- pervisory body for projects that do not fall within the jurisdiction of any existing body within the Department of Defense (DOD). The OSD (or its historical predecessor) has supported a variety of com- puter-related R&D efforts over the last several decades.) In the late 1950s, a DOD task force designed the specifications for Cobol, which ultimately became the standard language for business aml commer- cial applications. More recently' the DODi in';~ti~a\ted and supported the development of Ada, a programming language prompted by a defense-establishment-wide concern about the proliferation of di£fer- ent computer languages and the increasing dependence of the U.S. military on computers. In 1984, the DOD established the Software Technology for Adaptable Reliable Systems (STARS) program to pro- mote better software practice in both the military and the private sectors. Currently, the OSD (through the Office of the Director of Defense Research and Engineering) has begun to develop a software action plan to "develop and implement integrated technology and manage- ment plans to ensure more cost-effective software support."2 In con- junction with the management initiatives of this plan, the Software Technology Strategy is intended to reduce equivalent software life- cycle costs by a factor of two and to reduce software problem rates by a factor of ten by the year 2000, as well as to achieve new levels of mission capability.3 This strategy is based on five themes: software reuse, software reengineering to support already deployed systems, process support for software development, leverage of commercial technology for Defense Department needs, and the integration of ar- tificial intelligence and software engineering technology. Finally, over the last 40 years the National Security Agency (NSA) has played important roles in the development of supercomputers, primarily in support of its intelligence-gathering mission. NSA-re- lated research in CS&E has focused on high-performance computing, language processing, cryptography, and secure computing and com- munications. National Science Foundation Now the primary supporter of academic research in CS&E as measured by the number of individual investigators supported, the
222 COMPUTING THE FUTURE National Science Foundation (NSF) became a major supporter of CS&E research in the mid-1970s, when it shifted support for scientific ap- plications of computers to their parent sciences but left funding for the computer area unchanged, so that essentially the entire allocation became available for research in CS&E.4 By dollar volume, the NSF is now the second largest funder of CS&E research within the federal government. Figure 7.2 illustrates the NSF's history of funding CS&E research for the last 15 years. The budget for CS&E is the fastest growing budget category at NSF, although the budgets for other dis- ciplines start at much higher levels. Another major turning point in the relationship of the NSF to CS&E was the formation of the CISE Directorate in April 1986. Prior to 1986, CS&E received funding through several directorates (engi- neering, mathematics and physical sciences, and biological and be- havioral sciences). A memo to NSF staff from then-director Erich 150 140 130 120 1 10 cad 0 100 cad SO of' 80 70 60 50 40 30 20 10 O 1 , 1 , 1 To all recipients /,~' U~ To academia l / , 1 , 1 , 1 , 1 1 983 1 985 1 987 1 989 1 991 1977 1979 1981 Fiscal Year FIGURE 7.2 National Science Foundation obligations for research for com- puter science (basic and applied), FY 1976 to FY 1991, in constant FY 1992 dollars. SOURCE: Basic data (in then-year dollars) for all recipients and academia were taken from the corresponding sources cited in the caption for Figure 7.1.
INSTITUTIONAL INFRASTRUCTURE OF ACADEMIC CS&E 223 Bloch stated the rationale for restructuring: "NSF has considerable activities in computer science, information science, computer engi- neering, supercomputers and networking. Our investment in these new and important areas is growing rapidly. Many of the existing projects, programs, and initiatives are interrelated and support a com- mon community of scientists and engineers. In order to assure a broad and thorough understanding of our opportunities and respon- sibilities, a closer linkage between these organizationally separate groups is important."5 NSF's CISE Directorate is the primary federal supporter of inves- tigator-initiated CS&E research, although programs in other director- ates do support related research. For example, elements of the FY 1993 High Performance Computing and Communications Program, discussed in Chapter 1, can be found in the Biological Sciences Direc- torate, for protein folding; the Engineering Directorate, for optical computing; and the Mathematical and Physical Sciences Directorate, for parallel algorithms for computational physics. Prior to the formation of the CISE Directorate, the case for fund ing CS&E research was argued not by computer scientists or engi- neers but by others without substantial background in CS&E. Cur- rent and former NSF officials argue that the combination of several programs under the CISE Directorate strengthens the institutional influence of the CS&E community.6 In addition, the creation of the CISE Directorate is an acknowledgment that CS&E as a discipline is sufficiently different from others to warrant consideration on its own; this point echoes those made in the Chapter 6 section "Intellectual and Structural Characteristics of CS&E as a Discipline" about differ- ences between CS&E and other disciplines. Figure 7.3 illustrates various programmatic statistics of signifi- cance to the CS&E community:7 · The number of proposals submitted and awards made has grown steadily and substantially since FY 1986. However, proposal growth has outstripped award growth for most of the period from FY 1985 to FY 1990, leading to a declining success rate (i.e., the ratio of propos- als funded to proposals submitted). In FY 1990, the success rate rose for the first time in several years, from 26 percent in FY 1989 to 30 percent in FY 1990; it is now comparable to the average across all NSF directorates. (Nevertheless, CISE officials report that they re- ceive more scientifically meritorious proposals than they can fund. Several current and former CISE officials have said that their best guess is that on average, about 50 percent of proposals submitted would probably produce good science.)
224 2 1 .9 1 .8 1.7 1.6 1.5 1 .4 1.3 1 .2 1.1 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 O 1985 1986 1987 1988 1989 1990 COMPUTING THE FUTURE - - Relative Number ,~- of Proposals ~~ ,, , _ - Relative Award Size (in constant dollars) ~1 1 1 Fiscal Year FIGURE 7.3 Changes in program statistics for the NSF Computer and Infor- mation Sciences and Engineering Directorate, FY 1985 to FY 1990, including relative number and size of awards, and relative number of proposals sub- mitted. 1985 = 1.0. SOURCE: National Science Foundation, backup docu- mentation for "Background Material for Long-Range Planning: 1993-1997," National Science Board, NSF, Washington, D.C., June 20-21, 1991. · The constant-dollar value of the median award dropped by about 20 percent between FY 1985 and FY 1990, a trend that has raised concern in the community, given the increasing costs of coin research. The CISE Directorate allocates a little under 10 percent of its bud- get to the development of institutional infrastructure to support ex- perimental computer science and engineering ($19 million under the FY 1992 spending plan, out of a total CISE budget of $210.9 million); the impact of this program on universities is discussed below (see the section "Private Nongovernmental Organizations"~. A far larger portion of its budget (about 47 percent for FY 1992) supports a substantial computing infrastructure for use by the general science and engi- neering community as well as the CS&E field. The most important
INSTITUTIONAL INFRASTRUCTURE OF ACADEMIC CS&E 225 aspects of this infrastructure are the NSF supercomputer centers ($64.3 million), NSFNET ($25.8 million), and several science and technology centers ($8.8 million). NSF Supercomputer Centers The four NSF supercomputer centers provide academic and in- dustrial users with powerful state-of-the-art computational capabili- ties. These centers were established in 1985-1986; they are not now and have never been intended to be centers of CS&E research. But in the half-dozen years since their establishment, it has become increas- ingly clear that drawing sharp lines between providing computation- al facilities for other disciplines as opposed to CS&E is often unfeasi- ble. For example, as new parallel computers become available at the centers, nearby departments of CS&E may use them for educating their own students about new parallel programming paradigms. Given the increasingly varied choice of parallel architectures on the market, supercomputer centers and CS&E departments may find it beneficial to cooperate in choosing machines appropriate to the local environ- ment. In addition, it is true that most novice users are unable to exploit the full potential of supercomputers without extensive consultation with computer scientists and engineers who have a much keener un- derstanding of the hardware and software available. As these con- sultations have proceeded, inadequacies in existing tools (especially software) have been identified, and work has been undertaken to eliminate these inadequacies. Some nontrivial portion of such work has been nonroutine work that by any reasonable standard qualifies as research. For example, the supercomputer centers have played a major role in the development of scientific visualization, i.e., display- ing for human consumption many megabytes of data in a form that is quickly and easily understood. Performance evaluation of new su- percomputer architectures is technically demanding. To the extent that novel architectures for parallel processing will first come into scientific and engineering use at the supercomputer centers, their role in providing software to exploit these architectures will increase, requiring even greater CS&E effort to develop such software. Finally, the supercomputer centers are likely to serve an ever larger clientele in the future, most of whom will not have local ac- cess. Thus the centers may become hubs for high-speed networking activities that will require substantive CS&E input. The role of the supercomputer centers in technology transfer to the community at large has also increased as many of the software
226 COMPUTING THE FUTURE tools developed within these centers have been put to use in other high-performance computing environments. For example, these cen- ters have been major distributors of so-called coordination languages (e.g., Linda, char, Pam, and Xpress); such languages are integral to machine-independent programming environments that facilitate the transfer of programs between computers, ranging from networked workstations acting as a single machine to large-scale parallel ma- chines. Transfer of software by network (the "file transfer protocol," or FTP) accounts for a great deal of technology transfer. NSFNET The NSF also supports the NSFNET, the backbone of a network that connects hundreds of colleges and universities in the United States with high-speed links and is used by departments of all variet- ies, including CS&E. The extent to which NSFNET serves CS&E ver- sus other disciplines is unclear. Given the role that CS&E depart- ments have played in the development of network services and data communications, it is likely that CS&E department members use NSFNET more than members of other departments. Yet workers in other disciplines often need to transfer data in much larger quantities (e.g., for scientific visualization) than do computer scientists or engi- neers, and so CS&E may be a less data-intensive user than other disciplines. Science and Technology Centers Finally, in recent years, the NSF has begun to support interdisci- plinary science and technology centers (STCs). Three involve CS&E departments in a major way the STC for Discrete Mathematics and Theoretical Computer Science (involving Rutgers University, Prince- ton University, Bell Laboratories, and Bellcore), the STC for Comput- er Graphics and Scientific Visualization (involving Brown University, the University of Utah, Cornell University, the University of North Carolina, and the California Institute of Technology and partially supported by DARPA, IBM, Digital Equipment Corporation, and Hewlett- Packard as well as NSF), and the STC for Parallel Computing (in- srolving Rice University, the California Institute of Technology, Ar- gonne National Laboratory, Oak Ridge National Laboratory, and Los Alamos National Laboratory). The STCs are intended to support work on "complex research problems that are large-scale, of long duration, and that may require specialized facilities or collaborative relationships across scientific and engineering disciplines."8
INSTITUTIONAL INFRASTRUCTURE OF ACADEMIC CS&E National Aeronautics and Space Administration 227 The CS&E research of the National Aeronautics and Space Ad- ministration (NASA) involves concurrent processing, highly reliable cost-effective computing, scientific and engineering information man- agement, and artificial intelligence (AI). The first three areas support work in networked access, management of large scientific data sets, scientific visualization, massively parallel processing, development of very reliable, very complex software, and software producibility. The AI effort is relevant to a variety of NASA responsibilities and focuses on expert systems for diagnostic, consulting, and ultimately on-line control of shuttle and planetary probe operations, dynamic schedulers for shuttle operations, and large-scale capture of knowl- edge for use in knowledge engineering databases. NASA's support for CS&E has fluctuated cons~eraoty over the years, as has the fraction that has gone to universities and colleges. Figure 7.4 illustrates NASA's history of funding CS&E research for the last 15 years. _ _ . . . . 100 . 90 80 cn _ 70 Cal 60 50 11 o c - 30 ._ 20 10 To all recipients f I if\ \ / {I \ \ \ 1977 To academia ~ _0' / 1 1 1 1 1979 1981 1983 Fiscal Year - - 1 1985 1987 1989 1991 FIGURE 7.4 NASA obligations for research for computer science (basic and applied), FY 1976 to FY 1991, in constant FY 1992 dollars. SOURCE: Basic data (in then-year dollars) for all recipients and academia were taken from the corresponding sources cited in the caption for Figure 7.1.
228 COMPUTING THE FUTURE In the 1960s, the Apollo program made substantial use of ad- vanced computer systems.9 NASA focused on reliable and fault- tolerant computing. In the early and mid-1970s, NASA supported computer work related to the space shuttle, which declined as the shuttle reached operational capability in the early 1980s. NASA initi- ated work on the use of supercomputing for image processing and modeling of aerodynamic structures and created several centers (the Research Institute for Advanced Computer Science, the Institute for Computer Applications in Science and Engineering, and the Center of Excellence in Space and Data Information Studies) in which a sub- stantial amount of internal and external CS&E research is supported. In recent years, NASA has started to focus more on issues of scientif- ic data management, as the forthcoming Mission to Planet Earth be- gins. (More information on the computing aspects of NASA's Earth Observing System is contained in Chapter 2.) In addition to its support for CS&E research, NASA spends about $40 million per year on computational science and modeling and an additional $250 million per year on computational facilities (includ- ing networking, equipment leases, and software support); indeed, NASA spends more on supercomputers than does NSF, although NASA supercomputing is mission oriented, whereas NSF supercomputing serves many research users. Department of Energy The Office of Energy Research (OER) is the primary source of funding for CS&E research supported by the Department of Energy (DOE), which includes work on programming languages, automated reasoning systems, distributed systems, machine architectures for scien- tific computation, algorithms for parallel computing, and manage- ment of scientific data. Future programs are likely to emphasize distributed and massively parallel computing, portable and scalable libraries, environments for computational science, security, visual- ization and imaging, and very large scientific databases. Since 1945, the DOE and its predecessors have supported the development of high-performance supercomputers for their applica- tion in the design and development of nuclear weapons. Indeed, the first American electronic digital computer ever developed ENIAC- was used to support problems in computational physics and engi- neering associated with the development of atomic bombs in the post- war era.~° Along the way, a variety of supercomputer applications relevant to other DOE missions have emerged, and a great deal of sophisticated mathematical software has been distributed for general
INSTITUTIONAL INFRASTRUCTURE OF ACADEMIC CS&E 35 an ct 25 Cot 0' 20 15 o n o - 10 229 30 _ To all recipients / ~ To academia ,o--o ~ ~ I ol r I ~ I , I , I , I I - I,_ ~ ~ My / l ~-~-~ l 1 , 1 1 1 1987 1989 1991 1977 1979 1981 1983 1 985 Fiscal Year FIGURE 7.5 Department of Energy obligations for research for computer science (basic and applied), FY 1976 to FY 1991, in constant FY 1992 dollars. Basic data (in then-year dollars) for all recipients and academia were taken from the corresponding sources cited in the caption for Figure 7.1. use outside the nuclear weapons community. Given its interests in simulation, DOE has been an important stimulator of developments in high-performance computing and computational science, and it has provided an important market for the domestic supercomputer industry. The DOE is the fourth largest funder of CS&E research within the federal government. Figure 7.5 illustrates the DOE's history of fund- ing CS&E research from 1976 to 1991. A substantial fraction of the DOE budget for CS&E research is consumed by national laboratories, whose future with respect to budgets and shifts to civilian work after the collapse of the Soviet Union remains to be seen. Other Federal Agencies Other federal agencies account for only a small fraction of the total CS&E research budget. Among these, two are notable.
230 National Institutes of Health COMPUTING THE FUTURE The National Institutes of Health (NIH) has supported several important but specialized advances in computer science, particularly in expert systems for medical purposes. In the late 1960s, it took over funding for a former NASA project, DENDRAL an expert sys- tem developed to interpret mass spectrograms and thus to elucidate chemical structures. DENDRAL laid many of the foundations for current expert systems. In 1973 NIH began to support a center at Stanford University for applications of AI to medicine and biology. Work at this center has led to a variety of expert systems: MYCIN for matching patients with serious infections to appropriate antibiot- ics, PUFF for diagnosing lung diseases, the CASNET glaucoma spe- cialist, and INTERNIST, a diagnostic system for internal medicine. The NIH does not today support a great deal of research that it identifies as CS&E research per se. However, it does sponsor exter- nally and conduct internally a large amount of biomedical research that has important CS&E components. A small fraction of the total NIH budget for biomedical research of about $6.5 billion per year supports computational tools for medical research, mostly for soft- ware development. Computer science-related activities supported by the NIH include imaging and virtual-reality projects, molecular modeling, high-speed computing, large-database technology, statis- tics, instrumentation, AI and expert systems for medicine, medical language systems, and simulation. National Institute of Standards and Technology The National Institute of Standards and Technology (NIST) with- in the Department of Commerce houses the National Computer Sys- tems Laboratory, an in-house research effort in computer science with resources of about $25 million per year and 250 people, but does not support extramural research. NIST conducts some CS&E research (e.g., on optical character recognition) that is focused primarily on the needs of other government organizations and agencies. Never- theless, private industry makes considerable use of NIST work, since NIST plays a key role in setting standards and does other important work in security. The NIST also supports the Advanced Technology Program (ATP), a program to support the development of generic, precompetitive technologies. The ATE program was funded at $50 million for FY 1992 and is directed primarily at individual businesses or consortia of businesses and universities.
INSTITUTIONAL INFRASTRUCTURE OF ACADEMIC CS&E PRIVATE NONGOVERNMENTAL ORGANIZATIONS Universities 231 Universities and departments are a key aspect of the institutional infrastructure that supports academic CS&E. But the youth of CS&E as a discipline has led to certain anomalies in its role within the university. For example, in contrast to disciplines such as chemistry and physics that are overwhelmingly housed in departments dedicated to these disciplines and are generally located in colleges of arts and sciences, academic CS&E programs are housed in a variety of departments (Table 7.1~. Highly rated programs in CS&E are housed variously in autonomous departments (e.g., the Department of Computer Science at Stanford University), in mixed departments (e.g., the Department of Electrical Engineering and Computer Science at MIT and at the University of California, Berkeley), and in separate schools (e.g., the School of Computer Science at Carnegie Mellon University). The Computer Science Department at Browrr University is treated as any other department in a university of arts and sciences, whereas the Computer Science Departments at UCLA and the University of Penn- sylvania are located within the school of engineering; the Computer Science Department at Cornell University is part of the college of arts and sciences and the college of engineering. TABLE 7.1 Departmental Titles for CS&E Department Title Computer Science(s) Electrical and Computer Engineering Computer and Information Science(s) Computer Science and Engineering Electrical Engineering and Computer Science Electrical Engineering Computer Engineering Computing Science Computer Science and Operations Research Mathematical and Computer Sciences Other titles Number of Departments 92 19 10 13 10 2 4 2 2 3 9 (1 each) NOTE: A total of 166 departments are represented, out of a total of 168 Ph.D.- granting departments in the United States and Canada. SOURCE: David Cries and Dorothy Marsh, "The 1990-1991 Taulbee Survey," Com- puting Research News, Volume 4(1), January 1992, pp. 8 if.
232 COMPUTING THE FUTURE CS&E Ph.D. production is concentrated in a relatively few de- partmer~ts. The 12 top-ranked departments of 137 Ph.D.-granting computer science (note: computer science only) departments award- ed 233 doctorates in computer science in the 1990-1991 academic year, or 27 percent of all computer science Ph.D.s in that year; the 36 top- ranked departments accounted for 57 percent of the Ph.D.s awarded. The Ph.D.-per-departmer~t average of these 36 departments (13.7 per department) was well over three times that of the remaining 101 departments (3.6 per department. Major research institutions are also the most important undergraduate source for academic CS&E Ph.D. graduate students (Table 7.2~. The number and size of Ph.D.-granting departments in computer science have grown considerably in the past several years. Accord- ing to the annual Taulbee surveys, in 1984-1985 there were 103 such departments with a total of 1741 faculty members (or 16.9 faculty members per department); by the 1990-1991 academic year, these fig- ures had increased to 137 departments with 2725 faculty members (or TABLE 7.2 Baccalaureate Origins of Doctorate Recipients in CS&E, by Carnegie Classification, 1989 Carnegie Classification Computer Science Computer Engineering Research Ia Research IIb + Doctorate GrantingC Comprehensived Liberal Artse Other Total with known classification Total Ph.D.s 43% 24% 19% 11% 3% 291 531 50% 26% 12% 5% 7% 42 117 aUniversity receives at least $33.5 million per year in federal money for R&D and awards at least 50 Ph.D.s per year (e.g., University of California at Berkeley). bUniversity receives between $12.5 million and $33.5 million per year in federal money for R&D and awards at least 50 Ph.D.s per year (e.g., University of California at Santa Barbara). CUniversity awards at least 20 Ph.D.s per year in one discipline or 10 or more in three disciplines (e.g., University of California at Santa Cruz). dInstitution awards undergraduate and master's degrees only; more than 1500 stu- dents enrolled; more than half of undergraduate degrees awarded in occupational or professional disciplines (e.g., any university in the California State University system). eInstitution awards more than half of its degrees in liberal arts fields. SOURCE: Data from Survey of Earned Doctorates, Office of Scientific and Engineer- ing Personnel, National Research Council, Washington, D.C.
INSTITUTIONAL INFRASTRUCTURE OF ACADEMIC CS&E 233 19.9 faculty members per department).l3 (Figures for computer engi- neering for 1984-1985 are not available.) A second relevant aspect of university infrastructure is the capi- talization of CS&E departments. As noted in Chapters 1 and 6, re- search problems in CS&E are often driven and motivated by the up- per bounds of performance at the cutting edge of computing technology (whether the cutting edge results from sophisticated new components or novel arrangements of older components); good current examples include graphics and parallel computing. Research in computer graphics is very difficult today without the very fast graphics processors needed for three-dimensional displays, and experimental research parallel computing is impossible without access to parallel computers. How- ever, state-of-the-art systems are always expensive, and acquisition of such equipment does not benefit from the downward cost trend that characterizes computing equipment of a given sophistication or performance. Researchers in these areas are therefore often hard- pressed to assemble sufficient funds to pursue their research agen- das. Compounding the problem is the fact that a system that is state of the art today may not remain so for very long. Since hardware evolves rapidly, recently purchased hardware con- tributes more to the generation and solution of research problems than does older hardware. Since a considerable fraction of new CS&E Ph.D.s enter academia each year, and relatively few researchers re- tire, the pool of CS&E researchers competing for access to state-of- the-art equipment grows ever larger. Nevertheless, annual equip- ment-acquisition budgets remain level at best, and the trend indicated in Figure 7.6 suggests that annual spending on equipment has even begun to drop. The inescapable conclusion is that the availability of state-of-the-art computational resources is not keeping up with the demand for their use, and that this has been true for a long period of time. In addition, academic computer scientists and engineers have of- ten expressed concern that the costs of software are not adequately included in most assessments of capitalization. Software is of course a key element of research in CS&E, but the available data do not permit a determination of the extent to which software is included in assessments of capitalization. Capitalization for educational purposes is also an important as- pect of acquisition budgets. As noted in Chapter 1, students who must use computer systems with limited capability must often strug- gle with machine limitations rather than focusing on central concepts that could be more clearly illustrated with more powerful machines. For example, truly interactive visualization or computer-aided de
234 40 30 o COMPUTING THE FUTURE 80 70 60 20 ~ 10 _ O 1 1 1 1981 1982 1983 1984 - - - - - - 1985 1986 1987 1988 1989 1990 Fiscal Year FIGURE 7.6 Academic spending on equipment for use in computer science research, FY 1981 to FY 1990, in constant FY 1992 dollars. SOURCE: Basic data (in then-year dollars) provided by Science Resources Survey, National Science Foundation, Washington, D.C. sign (CAD) requires a response time of less than a few tenths of a second between user input and screen response. A visualization or CAD that responds in 2 seconds rather than 0.2 seconds gives the user an entirely misleading sense of its full value and potential. Equipment capitalization is concentrated in a relatively few de- partments. In 1988, 20 institutions had about 58 percent of the dollar value of computer science research equipment held by a total of 147 institutions (including those 20~;~4 these figures do not include com- puter centers operated for the benefit of the entire institution. The concentration of resources for CS&E research in a few select- ed institutions has been noted from time to time by the CS&E com- munity. For example, the Feldman reports issued in the late 1970s argued that experimental computer science was threatened by inade- quate equipment capitalization at too many schools. One response to these concerns was the Coordinated Experimental Research (CER) Program initiated in 1979 by the National Science Foundation. This program was designed to support the development of research equipment infrastructure at universities for the support of experimental research
INSTITUTIONAL INFRASTRUCTURE OF ACADEMIC CS&E 235 projects in computer science. Universities were selected on the basis of having strong CS&E programs.l6 The ultimate purpose of the CER Program and its follow-on (the Institutional Infrastructure Pro- gram) is to increase the number of universities that are capable of performing sophisticated experimental CS&E research (by faculty and graduate students engaged in dissertation work). Under the present Institutional Infrastructure Program, first-time awards range from $2 million to $4 million for five years, or about $400,000 to $800,000 per year; the FY 1991 budget allocated about $16.5 million to the Institu- tional Infrastructure Program. Efforts (including but not limited to the NSF CER Program) to support the research equipment infrastructure in CS&E have been largely successful. For example, 62 percent of the research equip- ment owned by the 127 other CS&E departments in 1988 was pur- chased in the two years preceding, compared to 52 percent in the top 20 departments.l7 However, unless a CER grant is renewed, grants terminate in five years, leaving recipients to pay afterwards for both replacement and maintenance. University infrastructure for CS&E may gain a further boost from the High Performance Computing and Communications Program. Of course, as previously noted, actual funding levels for the HPCC Pro- gram have yet to be determined. Professional Organizations Several professional organizations have had an impact on the practice of research and education in CS&E. These organizations include the Association for Computing Machinery (ACM) and the IEEE Computer Society, the Computer Science and Telecommunica- tions Board of the National Research Council, and the Computing Research Association. The ACM and the IEEE Computer Society are the leading profes- sional societies for CS&E. For example, the dozen or so publications each of the ACM and the IEEE Computer Society are major channels for the archival storage of new results and at times provide the first public look at innovations in commercial computing technology. Some of the journals published by these organizations are the most presti- gious in CS&E; others are sent to the entire membership of the orga- nization and thus serve to promote intellectual awareness of other subspecialties among more narrowly focused researchers. Both organizations also sponsor a wide variety of conferences and workshops every year. Conferences and workshops serve to disseminate new results more rapidly than is possible through print- ed media, a feature that is particularly important to a field as fast 1
236 COMPUTING THE FUTURE moving as CS&E. For the ACM, conferences and workshops often revolve around its 30 or so special interest groups (SIGs). The ACM SIGs are proposed, organized, and operated by a group of research- ers in a particular area of the field who want more interaction with their colleagues. Some of these SIGs are quite large and involve most of the important researchers in a given subspecialty. Several confer- ence proceedings (e.g., those of SIGGRAPH (SIG on computer graph- ics), SIGOPS (SIG on operating systems), SIGCOMM (SIG on net- working and communications), FOCS (foundations of computer science), SIGACT (SIG on automata and computer theory), SIGPLAN (SIG on programming languages), and SIGARCH (SIG on computer architec- tures)) are prestigious and tightly refereed; thus they often serve as the premier vehicles of dissemination for developments in the fields they cover, and are often preferred over archival journals. Conferences sponsored or organized by the IEEE Computer Soci- ety center on its 30 or so technical committees (analogous to the special interest groups of the ACM). Some technical committees are also quite large and have had a major impact on the field. The IEEE Computer Society has also played a role in the promulgation of stan- dards for various computing technologies. Undergraduate education in CS&E in its early days owes much to the ACM, which has been responsible for a number of initiatives over the years in developing curricula for undergraduate degrees in computer science. For example, the ACM sponsored the first major work on curricula in computer science, Curriculum 68, which had a major influence on the undergraduate curriculum in the many CS&E departments formed in the 1970s. More recently, the ACM and the IEEE Computer Society have worked together on curricular efforts, and they jointly created the Computer Science Accreditation Board, an organization that accredits undergraduate departments of com- puter science. For many years, academic CS&E lacked a major voice in the pub- lic policy debate. By contrast, most other disciplines have an organi- zation that represents that discipline to society. In many instances, policy makers know about these organizations and respect their judg- ments on issues of public importance. The organization monitors events, provides information when requested, organizes task forces on topics that need attention, keeps in touch with similar organiza- tions in neighboring fields, and works to inculcate in its members the idea that service to the community and society is not only useful but necessary. An example is the American Physical Society (APS), which repre- sents the research physicists of the nation. In the midst of the debate
INSTITUTIONAL INFRASTRUCTURE OF ACADEMIC CS&E 237 over the Strategic Defense Initiative in the late 1980s, the APS issued what was widely regarded within the public policy community as an authoritative report on the feasibility of directed-energy weapons for defense against strategic ballistic missiles. In recent years, two rather different organizations have begun to serve such a role for the CS&E research community. They are the Computer Science and Telecommunications Board and the Comput- ing Research Association. The Computer Science and Telecommunications Board (CSTB) of the National Research Council (the operating arm of the National Academy of Sciences, the National Academy of Engineering, and the Institute of Medicine) provides representation for the computing an communications field in a prestigious organization that provides in- dependent analysis and advice to the federal government. The char- ter of the CSTB is to examine technical, competitiveness, and public policy issues related to computer and communications science and technology. In this role, the CSTB composes study committees of leading computer scientists and engineers in academia and industry; convenes high-level meetings among senior researchers, executives, and government officials to discuss specific issues; and produces and disseminates reports. Through its activities, the CSTB promotes ac- tive intellectual cross-fertilization among the technical, business, and public policy communities. The Computing Research Association (CRA) is supported prima- rily by academic departments of CS&E that engage in research acti~r- ity, whether doctorate-granting or not, and engages the public policy process on their behalf. In addition to sponsoring the biannual Snowbird meetings of departmental chairs, the CRA is responsible for the an- nual Taulbee surveys of Ph.D.-granting departments. It also issues a well-received newsletter, organizes other surreys and reports where appropriate, and promotes service work in the CS&E community. Other professional organizations that serve the CS&E community are the Society for Industrial and Applied Mathematics (emphasizing the theory and computational aspects of CS&E), the Computer Pro- fessionals for Social Responsibility (an organization representing those interested in the social impact of computing technology), and the IEEE Communications Society (serving the networking community). NOTES 1. Kenneth Flamm, Targeting the Computer: Government Support and International Competition, The Brookings Institution, Washington, D.C., 1987, pp. 75-76. 2. U.S. Department of Defense, Department of Defense Software Technology Strategy, December 1991, prepared for the Director of Defense Research and Engineering, p. ES-1.
238 COMPUTING THE FUTURE 3. U.S. Department of Defense, Department of Defense Software Technology Strategy, 1991, p. ES-2. 4. Kenneth Flamm, Targeting the Computer, 1987, p. 88. 5. John Walsh, "NSF to Establish Computer Directorate," Science, Volume 232, April 4, 1986, page 18-19. 6. This strength is illustrated by the fact that the CISE Directorate's budget has grown significantly relative to those of the other research directorates within NSF (from 8.5 percent of the total NSF budget in FY 1986 to about 11 percent in FY 1991). Moreover, although the CISE budget provides for service functions to the entire sci- ence and engineering community as well as research support for the CS&E community (e.g., the NSF supercomputer centers and NSFNET), the research component of the CISE budget exhibits a similar trend. Put another way, growth in the service functions of the CISE directorate is not disproportionately responsible for growth in the overall CISE budget. An easily available source for the funding history of NSF and CISE can be found in Terry Walker, "A Review of Federal Funding for Research in Computer Science and Engineering," Computing Research News, April 1990, pp. 6-14. 7. Data presented for FY 1985 and FY 1986 are for those proposals submitted to the various NSF programs that were consolidated into the CISE Directorate in 1986. 8. National Science Foundation, NSF Science and Technology Research Centers, OMB 3145-0058, undated. 9. Kenneth Flamm, Targeting the Computer, 1987, pp. 84-85. 10. Kenneth Flamm, Targeting the Computer, 1987, p. 78. 11. Kenneth Flamm, Targeting the Computer, 1987, pp. 90-91. 12. David Gries and Dorothy Marsh, "The 1990-1991 Taulbee Survey," Computing Research News, Volume 4(1), January 1992, pp. 8 If. 13. Data for 1984-1985 are taken from David Gries, "The 1984-1985 Taulbee Sur- vey," Communications of the ACM, Volume 26(10), October 1986, pp. 972-977. Data for 1990-1991 are taken from David Gries and Dorothy Marsh, "The 1990-1991 Taulbee Survey," Computing Research News, Volume 4(1), January 1992, p. 10. 14. These 20 institutions have about $97.9 million of total in-use research equipment held by all institutions in the sample. This estimate is derived by multiplying the mean dollar amount of computer science research equipment for these 20 institutions (listed in Table 7 in the NSF report cited below as $4.895 million) by 20. Table 2 in the same report lists the aggregate purchase price of research equipment in these institu- tions as $168 million. See National Science Foundation, Academic Research Equipment in Computer Science, Central Computer Facilities, and Engineering: 1989, NSF 91-304, NSF, Washington, D.C., 1989, Table 2 (p. 4) and Table 7 (p. 7). 15. Jerome A. Feldman and William R. Sutherland, "rejuvenating Experimental Com- puter Science," Communications of the ACM, September 1979, pp 497-502. 16. The three institutions with the largest federal grants for computer science (Stan- ford University, Carnegie Mellon University, and MIT), each with about $5 million to $8 million annually in federal funding for computer science in 1979, agreed not to apply for these grants. 17. See National Science Foundation, Academic Research Equipment in Computer Sci- ence, Central Computer Facilities, and Engineering: 1989, NSF 91-304, NSF, Washington, D.C., 1989, Figure 4, p. 8. 18. Maintenance and repair costs are considerable. For example, annual expendi- tures for equipment maintenance and repair are about $0.37 per dollar of CS&E re- search equipment, compared to an average of $0.21 per dollar of scientific and engi- neering equipment taken across all fields. Indeed, CS&E maintenance and repair costs are the highest among those for all science and engineering fields. See National Sci- ence Foundation, Academic Research Equipment and Equipment Needs in Selected S/E Fields: 1989-1990, NSF 91-311, NSF, Washington, D.C., May 1991, Table 3, p. 4.
