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.
1 Computing Significance, Status, Challenges COMPUTING IN SOCIETY Computing is inextricably and ubiquitously woven into the fabric of modern life. In nearly all sectors of the economy, computing makes it possible to deliver services and products of higher quality to more people in less time than would otherwise be possible (Box 1.1~. As seen from the perspective of other technical fields (Box 1.2) and in terms of its potential to enhance U.S. industrial strength and the national defense (Box 1.3), computing is a truly enabling and central technology. Consider: · In large businesses, electronic mail enabled by computing is increasingly common. · In communications, computing makes it possible to switch and route over 100 million long-distance telephone calls per day. · In aeronautics, computer-aided design techniques are expected to save Boeing as much as a billion dollars in the development of the 777 airliner. · In pharmaceuticals, computing enables chemists to conduct sys- tematic searches for compounds that will fight specific diseases. · In automobile engineering, computing makes it possible to simulate automobile crash tests that would otherwise cost hundreds of thou- sands of dollars apiece.2 · In the oil industry, computing has saved hundreds of millions 13
14 COMPUTING THE FUTURE . ......................................... ..... . . .. .. .. . . . . . . . . . . . . . . ............................................................. T ~.~. ~.~. ~,£.~.~.~.~ . . : ... ..... ... .. ....... . . . . . .. . . .: ·, _"''a"""~"""~.," ''''' ''''"'''''"''''' '''''''''' '' ''"'''""'""''''' .............. i. . . . . . . . . . . . . . . . ~.. A ......... ........... . . mt er ~: ~l Me WAS :0t=~:r:~mo~.~: ~ :~- ................... . . vI~61~ 1,=3.te...~t::c~0 per '~8 task ma| ., . . .. , . .. .. , .. - . ... . . . . . . C.~.. pew - p<>;~sal:~ ]1~ `~t=~\Q mat '.'.~'. "'''c""a'.""""'""' "' ''' ' ' ' ' " ' ' ' " '' ' "''' "" ' '" ''' ' ' "" ... . . . . .. . .. .. . . . .. . . . .... . . . i. * . . .. . . ~ - ~ i: ::: :: : :: ::::: :: : ::::: : :: : ~ : "''''.'.'.2. K '' . '' -' ' ' ' . ..... . . . ....... '.' ' ' ' '." 2 ' 2.,' ' ' .,, . '. ' .''.' .. ...... .... '' '''"''''"2m2'"a '' ' ''' '' ' ''' ' "' " " ' ' '' ' " ' ' ' ' ''' ' ' ' "' ' ' ' 'I' '"" '' "' ' "' ' ' '' ' ' ' ''" '' ' "' '. '2'.. 't"2''8"t','l''''''""""' ' '"" " " ' " .''""" """""""""""'"''""'": .''"'"' '''"'''"", """""""''""""" " """: :':"".""""''."" "'.'" "" "'""'"""""""'.' """.""'."'." "" te _ ~y~ss~eifl~ pta8~!l~ ~ ~6 ~e5~e ................................................................................................................ eme noes tn compu~r s'mu~ ~S 8~t t~t t0~?Y . ... ~ . ~l,~r,~ ....''.~"""'"~.,'" ' ''"""' ' ' " ' '' '' ' ' ' '"' ' ' ' '' ' ' '' ' ' '' ' ' ' ' ' ' ' ' "'" : ::: :: :: : : : : ::: :::; : :t :: : :: :: : : :::::::::::: : ::: ::::::::::: :::::: ::::::::::::: :::::::::::::: ::::::::::::::: ::: :: :: :::: ::: .::::::::: ::::: . . h ~ ::::::: ::::::::::: :: ::::: : :::: : :: :: : : : : :: : . .. ~ : :: i: : ::::::: :: ~ :::::: ::: ~:: ::: ::::::::: :::: :::::::: ::::: :: :: ::: ::::: :::: :::::: :::: ::: of dollars in the past five years by helping drillers to avoid "dry" wells.3 · In offices, computer-based spreadsheets enable thousands of analysts and managers to model and predict financial and economic trends. · In science, computing is becoming a third paradigm of scien- tific inquiry, on a par with theory and observation or experiment and often yielding unexpected or unanticipated insights not possible through purely theoretical or experimental means.
COMPUTING-SIGNIFICANCE, STATUS, CHALLENGES 15
16 COMPUTING THE FUTURE
COMPUTING-SIGNIFICANCE, STATUS, CHALLENGES 17 Why should computing be so important even essential in these and so many other areas of human endeavor? Fundamentally, the answer is that computing can be usefully applied to any endeavor that uses or can be made to use information in large quantities (~nfor- mation-plenty) or information that has been highly processed and manipulated (information-rich). Information-rich and information-plenty endeavors primarily in- volve products of the human mind- numbers, pictures, ideas. As a device that excels in the storage and manipulation of information, the computer serves primarily as an amplifier of human intellectual capabilities. By operating very rapidly, it enables ~nformation-plenty activities. By undertaking efforts that are beyond the intellectual reach of human abilities, it enables information-rich activities. It is this enabling amplifier effect that is at the heart of today's information revolution, a revolution that may be as significant to human destiny as the agricultural and industrial revolutions. To paraphrase John Seely Brown, corporate vice president of the Xerox Corporation, mass and energy are being replaced by information and computing. The examples above include vignettes on how comput- ing makes automobiles more energy-efficient and manufacturing less materially wasteful. But what is obvious only at a macro-level is the change in the national economy itself. Once buttressed primarily by the sales of material artifacts such as inventory parts, airplanes, and automobiles that derive their value from structuring the atoms that give them substance, the economy is now increasingly one of infor- mation artifacts that may, for example, derive value from structuring musical notes into a symphony, words into a book, binary digits into a computer program, or figures from a business projection modeled on a spreadsheet. Nowhere is the shift from tangible artifact to information artifact better illustrated than in the computer industry itself. In its first few decades of existence, the computer industry made its money in the manufacture of computers. Today, the software sector is the most rapidly growing and profitable sector of the industry, as illustrated by its 19 percent growth rate in 1990 over 1989 levels versus a 9 percent growth rate in the industry overall.4 Yet software itself con- sumes no material, weighs nothing, and requires essentially no pow- er.5 Software is information crystallized in a particular form, and in this form it is valued at over $20 billion per year by the United States- and this estimate excludes the substantial amounts of custom software developed "in house" by computer users. Other examples of the increasing importance of information include the entertainment in- dustry (over $12 billion in sales in 1990 by five major entertainment
18 COMPUTING THE FUTURE companies6 and videogame manufacturers7 ~ and telecommunications ($107 billion in sales of services by the telephone companies listed in The Business Week 10008), both industries that trade mostly in ideas, information, and imagination. Information technology (including com- puter and communications hardware, plus computer software and services) directly accounts for around 5 percent of the GNP,9 even disregarding its enabling role in other sectors of the economy. SCOPE AND PURPOSE OF THIS REPORT -or ^ L~^~_~V -^ ~rr~ As a key force driving the development of ever more sophisticat- =~1 rmn~miltin~ once tic the cil~nlier of a large proportion of the trained computing personnel in industry, academic computer science and engineering (CS&E) has had a substantial impact on the nation.~° But today, both the intellectual focus of academic CS&E and the envi- ronment in which academic CS&E is embedded are in the midst of significant change. The intellectual boundaries of academic CS&E are blurring with the rise of in-depth programs and activities in com- putational science-the application of computational techniques to advance such disciplines as physics, chemistry) biology, and materi- als science. Universities themselves are retrenching; the computer industry is undergoing substantial and rapid restructuring; and the increasingly apparent utility of computing in all aspects of society is creating demands for computing technology that is more powerful and easier to use. In light of these changes, the Committee to Assess the Scope and Direction of Computer Science and Technology was convened to de- termine how best to organize the conduct of research and teaching in CS&E for the future. The result of its two-year study is an action plan that calls both for sustaining traditional core activities within CS&E and broadening the scope of CS&E's intellectual agenda as the field evolves into the 21st century. This report is divided into two parts. Part I addresses in broad strokes the fundamental challenges facing the field and discusses what the committee believes is an appropriate response to these challeng- es Chapter 1 briefly discusses the intellectual nature of CS&E and then elaborates on the nature of the impending challenges. Chapter 2 provides the philosophical underpinning for an appropriate response by the academic CS&E community. Chapter 3 outlines a core re- search agenda to carry CS&E into the future. Chapter 4 discusses the state of CS&E education at all levels. Chapter 5 articulates a set of judgments and priorities for the field and presents recommendations informed by those judgments and priorities. Part II explains in greater
COMPUTING- SIGNIFICANCE, STATUS, CHALLENGES 19 detail three aspects of the field: CS&E as an intellectual discipline, in Chapter 6; the institutional infrastructure of academic CS&E, in Chapter 7; and the demographics of the field, in Chapter 8. COMPUTER SCIENCE AND ENGINEERING Computational power however measured has increased dra- matically in the last several decades What is the source of this in- crease? The contributions of solid-state physicists arid materials scientists to the increase of computer power are undeniable; their efforts have made successive generations of electronic components ever smaller, faster, lighter, and cheaper. But the ability to organize these compo- nents into useful computer hardware (e.g., processors, storage de- vices, displays) and to write the software required (e.g., spreadsheets, electronic mail packages, databases) to exploit this hardware are pri- marily the fruits of CS&E. Further advances in computer power and usability will also depend in large part on pushing back the frontiers of CS&E. Intellectually, the "science" in "computer science and engineer- ing" connotes understanding of computing activities, through mathe- ~matical and engineering models and based on theory and abstrac- tion. The term "engineering" in "computer science and engineering" refers to the practical application, based on abstraction and design, of the scientific principles and methodologies to the development and maintenance of computer systems be they composed of hardware, software, or both. Thus both science and engineering characterize the approach of CS&E professionals to their object of study. What is the object of study? For the physicist, the object of study may be an atom or a star. For the biologist, it may be a cell or a plant. But computer scientists and engineers focus on information, on the ways of representing and processing information, and on the machines and systems that perform these tasks. The key intellectual themes in CS&E are algorithmic thinking, the representation of information, and computer programs. An algo- rithm is an unambiguous sequence of steps for processing informa- tion, arid computer scientists and engineers tend to believe in an algorithmic approach to solving problems. In the words of Donald Knuth, one of the leaders of CS&E: CS&E is a field that attracts a different kind of thinker. I believe that one who is a natural computer scientist thinks algorithmically. Such people are especially good at dealing with situations where different rules apply in different cases; they are individuals who can
20 COMPUTING THE FUTURE rapidly change levels of abstraction, simultaneously seeing things "in the large" and "in the small.''l2 The second key theme is the selection of appropriate representa- tions of information; indeed, designing data structures is often the first step in designing an algorithm. Much as with physics, where picking the right frame of reference and right coordinate system is critical to a simple solution, picking one data structure or another can make a problem easy or hard, its solution slow or fast. The issues are twofold: (1) how should the abstraction be repre- sented, and (2) how should the representation be properly structured to allow efficient access for common operations? A classic example is the problem of representing parts, suppliers, and customers. Each of these entities is represented by its attributes (e.g., a customer has a name, an address, a billing number, and so on). Each supplier has a price list, and each customer has a set of outstanding orders to each supplier. Thus there are five record types: parts, suppliers, custom- ers, price, and orders. The problem is to organize the data so that it is easy to answer questions like: Which supplier has the lowest price on part P?, or, Who is the largest customer of supplier S? By cluster- ing related data together, and by constructing auxiliary indices on the data, it becomes possible to answer such questions quickly with- out having to search the entire database. The two examples below also illustrate the importance of proper representation of information: · A "white pages" telephone directory is arranged by name: knowing the name, it is possible to look up a telephone number. But a "criss- cross" directory that is arranged by number is necessary when one needs to identify the caller associated with a given number. Each directory contains the same information, but the different structuring of the information makes each directory useful in its own way. · A circle can be represented by an equation or by a set of points. A circle to be drawn on a display screen may be more conveniently represented as a set of points, whereas an equation may be a better representation if a problem calls for determining if a given point lies inside or outside the circle. A computer program expresses algorithms and structures infor- mation using a programming language. Such languages provide a way to represent an algorithm precisely enough that a "high-level" description (i.e., one that is easily understood by humans) can be mechanically translated ("compiled") into a "low-level" version that the computer can carry out (unexecuted; the execution of a program
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 21 by a computer is what allows the algorithm to come alive, instructing the computer to perform the tasks the person has requested. Com- puter programs are thus the essential link between intellectual con- structs such as algorithms and information representations and the computers that enable the information revolution. Computer programs enable the computer scientist and engineer to feel the excitement of seeing something spring to life from the "mind's eye" and of creating information artifacts that have consid- erable practical utility for people in all walks of life. Fred Brooks has captured the excitement of programming: The programmer, like the poet, works only slightly removed from pure thought-stuff. He builds castles in the air, creating by the exertion of the imagination.... Yet the program construct, unlike the poet's words, is real in the sense that it moves and works, pro- ducir~g visible outputs separate from the construct itself.... The magic of myth and legend has come true in our time. One types the correct incantation on a keyboard, and a display screen comes to life, showing things that never were nor could be.~3 Programmers are in equal portions playwright and puppeteer, working as a novelist would if he could make his characters come to life simply by touching the keys of his typewriter. As Ivan Suther- land, the father of computer graphics, has said, Through computer displays I have landed an airplane on the deck of a moving carrier, observed a nuclear particle hit a potential well, flown in a rocket at nearly the speed of light, and watched a com- puter reveal its innermost workings.~4 Programming is an enormously challenging intellectual activity. Apart from deciding on appropriate algorithms and representations of information, perhaps the most fundamental issue in developing computer programs arises from the fact that the computer (unlike other similar devices such as non-programmable calculators) has the ability to take different courses of action based on the outcome of various decisions. Here are three examples of decisions that pro- grammers convey to a computer: · Find a particular name in a list and dial the telephone number associated with it. · If this point lies within this circle then color it black; otherwise color it white. · While the input data are greater than zero, display them on the screen. When a program does not involve such decisions, the exact se
22 COMPUTING THE FUTURE quence of steps (i.e., the "execution path") is known in advance. But in a program that involves many such decisions, the sequence of steps cannot be known in advance. Thus the programmer must an- ticipate all possible execution paths. The problem is that the number of possible paths grows very rapidly with the number of decisions: a program with only 10 "yes" or "no" decisions can have over 1000 possi- ble paths, and one with 20 such decisions can have over 1 million. Algorithmic thinking, information representation, and computer programs are themes central to all subfields of CS&E research. Box 1.4 illustrates a typical taxonomy of these subfields. Consider the subarea of computer architecture. Computer engineers must have a basic understanding of the algorithms that will be executed on the computers they design, as illustrated by today's designers of parallel and concurrent computers. Indeed, computer engineers are faced with many decisions that involve the selection of appropriate algo- rithms, since any programmable algorithm can be implemented in ... .. . . . ....... ... .. .... .... .. . ...... . ., ~.~.,,1.,. ~. XoN. ~.~.~. ~.~.~.~. ~. ~.~. : . . ..... .............................................. ~ ^~s and dam struct~ . ~.................... . ~.............. ................... .......... . .. .. . . ... * .. ....... 1 ... . . . ... .. . . ~ ,., ~.~.~. ~. ~I - ..... .... ~ . ~.. .. , ~.~. ~.~.~.~. ~. .... - ~ - -. ..... . . ~.~.~.e r] ( ~n d sy.~.~.~ ~.~.~.~.~ 5 oPerating~s~ms~ ~ ~ ~ ~ ~ ~ ~ ~* q ~= :: :: : Em: : : ~ :: :: :: : :::: ::: :: ::::: :::: ::: : :::::::: ::: ::: ~ -f ~ ~ ~ It~ . ~ - 1 . : ::: ::: . . :: ::::::: :: ::: :.: ::: ::::: :: :: :::: ::: ::.::::: ::::::: ::::: :::::: :: :::: :: ::: :::: i:: .. .. . .. ~ ~ ~ ~ ~ Add ' '."2d''e"''''' '' " ' '"'' ' ' '' ''' ' " ' '' '''" ' ':"'" ' '" ' ' ' '" ' ' ' ' ..... ......... ..... .... ...... .. . ~.t t ~ ~ ~ d [* ~ ~ ....'l.''"s'.'."'b'''' ' """ ""'''"'''""' '' "'' ' '' ' '"" ' ' ' ' " ' :'' "''' ' ''' ''' '' ' ' ' ' '' ' "': .............................. .. .. ... .... .... .. . ~.,.,~.k t., ~.~. ~. ~. ~. ~.~. ~.~.~.~. ~. ~.~ ~I . . ....... ...... ... . . . . ... . . . ~^.t,=~.t,.~. ~.: ~.~. ~: ~ . a. ~m .... ~ ~. ~.~ ~ ~ ~: ~ . . ~ ~ J . . : : :. : ,:., ,,, : : ~ : : : :: : : ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ! . ~ ~. ~, ~, ~,l. .. - ... . .............. . . . ... . . ash She," ~
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 23 hardware. Through a better understanding of algorithms, computer engineers can better optimize the match between their hardware and the programs that will run on them. Those who design computer languages (item two in Box 1.4) with which people write programs also concern themselves with algorithms and information representation. Computer languages often differ in the ease with which various types of algorithms can be expressed and in their ability to represent different types of information. For example, a computer language such as Fortran is particularly conve- nient for implementing iterative algorithms for numerical calcula- tion, whereas Cobol may be much more convenient for problems that call for the manipulation and the input and output of large amounts of textual data. The language Lisp is useful for manipulating sym- bolic relations, while Ada is specifically designed for "embedded" computing problems (e.g., real-time flight control). The themes of algorithms, programs, and information representa- tion also provide material for intellectual study in and of themselves, often with important practical results. The study of algorithms with- in CS&E is as challenging as any area of mathematics; it has practical importance as well, since improperly chosen algorithms may solve problems in a highly inefficient manner, and problems can have in- trinsic limits on how many steps are needed to solve them (Box 1.5~. The study of programs is a broad area, ranging from the highly for- mal study of mathematically proving programs correct to very prac . . . . . . .. . . ..... . i .. .... .. .... ~ . . ..~ ~.. .~ . ~ ~
24 COMPUTING THE FUTURE tical considerations regarding tools with which to specify, write, de- bug, maintain, and modify very large software systems (otherwise called software engineering). Information representation is the cen- tral theme underlying the study of data structures (how information can best be represented for computer processing) and much of hu- man-computer interaction (how information can best be represented to maximize its utility for human beings). CONTRIBUTIONS OF CS&E TO COMPUTING PRACTICE CS&E research has made enormous contributions to computing practice. Insights from CS&E research inform the approach of pro- grammers and machine designers at all levels, from those designing a still-faster supercomputer to those programming a small personal computer. Techniques and architectural themes developed or codi- fied under the banner of CS&E are familiar to every developer of software and hardware. Consider modern word-processing systems, familiar to millions of office workers with no technical training. Many features that make these systems so popular (e.g., full-screen "what you see is what you get" (WYSIWYG) editing, automatic line-wrapping at the end of a line, automatic pagination, mouse pointing) first appeared in text editors developed by computer scientists and engineers. As impor- tantly, the internals of modern word-processing systems depend on a host of algorithms and data structures investigated in the course of CS&E research: automata theory, dynamic programming, constraint satisfaction, incremental updating, partial-match retrieval, data com- pression. Spreadsheets, though not first conceptualized by computer scientists, also depend on many of these algorithms, data structures, and concepts for efficient implementation on personal computers. These ideas the result of CS&E research and disseminated by CS&E edu- cation are second nature in programming, just as Kirchhoff's laws, amplifiers, and flip-flops are elemental ideas in electrical engineer- ing. From only the most rudimentary idea of a word processor or spreadsheet, good programmers can quickly determine how to make one and can explain the plan concisely. Modern database management systems, for mainframes and per- sonal computers alike, rely on computer science and engineering re- search from top to bottom. For example, computer science research- ers in the late 1960s and early 1970s created the relational data model to represent data in a simple way. Computer engineers worked through the 1970s on techniques to implement this model. By the mid-1980s these ideas were understood Tell enough to be standardized by the
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 25 Interr~ational Organization for Standardization (ISO) in the language SQL. SQL has become the lingua franca of the database business. The committee estimates that today about 100,000 computer programmers in the United States use a database system as their main tool; hiring these programmers costs about $10 billion per year. Improving their productivity by even a small amount has a huge payoff, and most studies indicate that the relational database model and its associated tools more than double programmer productivity. CS&E has been profoundly helpful to much of modern science and engineering. For example, the speed with which certain types of partial differential equations may be solved has improved by a factor of around 10~i since 1945 (Figure 1.1), due in about equal measure to faster machines developed by computer engineers and better algo- rithms developed by mathematicians and theoretical computer scien- tists. Just as importantly, computer scientists have developed pro- grammir~g languages that enable scientists to use computers more effectively and computer-based techniques for interactive scientific 12 109 o C' IL 1 o6 a) a) Q En 103 10° Total Speedup r - it, .. - ~ Hardware ., Algorithms , ~,, CL -- 1950 1960 1970 1980 -Second -Minute -Hour . - -Week -Year -Century FIGURE 1.1 Speedup in the solution of Poisson's equation, on a grid 64 points on a side. Note that the increased speed results from both better computer hardware and better algorithms to solve the equation. SOURCE: Jon Bentley, More Programming Pearls, Addison-Wesley, New York, 1988, p. 158. Copyright @) 1988 by Bell Telephone Laboratories, Inc. Reprinted with permission of Addison-Wesley Publishing Company.
26 COMPUTING THE FUTURE visualization, in which huge amounts of data perhaps generated by solving these partial differential equations are transformed into easily understandable images. Indeed, visualization is now a new para- digm for the presentation of information. Many operating systems (i.e., the system software that provides basic machine functions on which applications software can build) such as MS-DOS and Unix make use of many years of experimental CS&E. Operating systems provide abstractions and comprise com- ponents developed through the study of CS&E: processes, files, ad- dress spaces, concurrency, synchronization. Various parts of operat- ing systems are engineered according to analyses of strategies for memory allocation, scheduling, paging, queuing, and communica- tion. In each case, the best modern practices to which implementors instinctively turn have been explored and codified by CS&E. Theoretical computer scientists studying computational complex- ity with mathematical tools have had a major impact on computer security. Modern cryptographic methods (e.g., public-key encryp- tion systems) devised to protect and guarantee the integrity of elec- tronically transmitted messages are based on work in complexity the- ory performed since the 1970s (Box 1.6~. As the complexity of computing has grown since the invention of the digital computer in the 1940s, so also has the need for well-un- derstood concepts and theories with which to manage this complexi- ty. Whereas intuitively grounded insight was often sufficient to lead to substantial progress in the earliest days of the field, a systematic approach has become increasingly important. Thus the importance of CS&E to computing can only be expected to grow. COMPUTING AS A TWO-EDGED SWORD - As with many technologies, the initial applications of computing to various problems were widely regarded as positive. This history has raised social expectations with respect to computer technology, ex- pectations that simply result in a larger "fall from grace" when the inevitable problems associated with any technology become mani- fest. New technologies widen the number of options with which a given problem may be addressed. But whether any particular option is desirable is an issue that depends very much on the perspective of the person considering it. A national point-of-sale networki5 might be highly desirable for merchants, since it could greatly reduce the need for cash and paper work, but undesirable for individuals wish- ing to keep private their buying patterns. A higher degree of auto
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 27 i. .. ~ ~ .. ~ . ~ i i. ~ ~ ..~ . ~ .~ ~ . ~ ~ . ~. .... i ~ . . ~ ~ .............. .. i. ~ . i ................... ., ., ,, 8~x . ~c.~.~. p .,. . . ~t . . I. : . . ~ ~ . . :. . : . . . : . ~, , . ~ ~............ ..................... ., ', ''''3"d'.2''2~,''''''2a2' ' ' '' "''' ' ' '' " '" ' ' ' ' ' ' ' '' ' ' ' '' '' ' ''" ' "' ' ' ' " ' '' '' ,. ~. . .............. m - Ace. etw - ~.them. nOwew(''''''p0'bt'i'c''Le'""'"'c'ry^s'smm's'""'el"~"~""'"' . ...................................... ............. . ...... .... ... j . .. f . . """'m"23""~"""" ' ' ' " ' ' ""'' '' ' ' ' ' ' '' ''' ' ' ' '' ' ' ' ' ' " '' " ' '' " '' ' ' "' '"' ' ' . i ~ . . . I ~ i, .. i ................... ~ ... . . , . . x . ......... ~ o Ace. parI? .~; lout. prolecIlo. ~. ~. ~. ~.n, ~ puO .. ~s nec ...f...~. .... . . ..~ . . ~ . . . . _~s~ - >sect - id t3$~9~tc kin_ .. ~ . ~ .~ ....................... ~ . ~ . . . . . ~* .... . ~; ~n eaw'suro:~r' wIl~{ access ~ Aim:: ^:~e'::s ::::: >Q )"s s~t ke . '. .'..'..."'"""'" "' ' '"" ' ' '' '"' ' "'""""'' """' '"' "" '"'''' '"'"" "'""" ' '"'""''"'"''' ''" ' ' '" "'''' "' ' ..... ~ , ...................................................................................................... ~ ~-'_~g ............... ,.,,l,l.,,, ,k,,,. ~.. . . . . . . . : ....... . .... ..................... . ~. .. .. a. . , ~. :: .:::.: :: :: : :: ::: h : ::: : : : ................... . .................................................. ~ .. ~ ........................ ,,,,,: 9~ ~t - ~'''^t'i'~=t~',''~''''~l''~'''~'''''~t''' jl~ `'''~' A'''' ' I''' ''I'''' ' '' '''' lid' '':''''':'' ''"'"''""""""""""''""""'"''T' "' " "' '''W'T"""'~'T' i':"""'," " '..''..".."""..""'..'.' .'T' """~"""~'T,.,.'~'T'''':~' ~ :~` ,.,,, ca.n,,~e ~ ~ ~t messa~ ct81.~.~.~.g...~.~.~..~...~..~.~........................... .,, . . ~. ~,. i,., ~l, ,~, .. .. . '''~'''''''mer,~} operM'~$ thet 3~e 6~t to ~ Or exa.~.~. ~. ~.~ .~ i. i . ~ . ~ ~ . ~ ~add ~ ~ nIn~bem 4439 are ~ ~ ~ ~ ~: d~r=~ beat :439: x163 1 t ............................... .. ' '. ' ~, ~? ' 09' " TheomtI'ca'' 'cOm9'uter"'scl'enttm hew played' ~ ma)~r m! "en ' ..... m.~.~.~. ~.~e, ~. ~. ~ coma mu. ...................................... b. ~h ~( .............. ......... ''.. '.'.''''p'""''"'''""''' ''' ' '"' ''' " ""a'' "' ' '' "'" ''''"' ''''"' "' '""' " "'' ' '"' "" ' ' '' '"""' '"""'" .''. '. ..'" ' ' '' " ' ' ' '' ' ' '"" ' ' " ' ' ' ' ' ' ' '' ' '' ""' '' ' "' ' ""'' '" ' '"''''' '"'"'""'' ' ''"' ' ' " ' ' "' " '' " ' '" ' ' "'' '"" ''I' "' .~ ................ ..... . it. ~ . .. . ~ . ~ ~ . i. .............. ............................................................................................................................ ... . . . . . . . .. . ............. .......... ....... ~ . ~ ....... : . ~. ................ .. ,..,,,,,........... .............. ............. .....~..~. .............. ............ ............ ............... ............. .............. ...~.......... .............. ................ . ......... ............... ..:...~......... ............. ............. ................ . . ..... .... .~......... · .....~... :::: :. : . :: . . ... . : :.. ~ : . :: . ::: : mation in an office may be good from the standpoint of office manag- ers wishing to increase efficiency, but bad from the standpoint of the union that fears a loss of jobs. Networked computers, introduced to facilitate computer-to-computer communication, now provide path- ways for computer worms and viruses, as the Internet "worm" inci- dent demonstrated in 1989. Moreover, the introduction of new technologies increases the number
28 COMPUTING THE FUTURE of things that can go wrong. An airplane simulator can have incom- parable value for training a pilot, but a design error in the simulator that results in a mismatch between simulated and actual airplane behavior could have disastrous results. The much-reported failures in 1991 of various new telephone switching systems, introduced to provide new telephone services, are another example. Early but er- roneous computer-based forecasting of election results could subvert the democratic process. False computer-generated reports of incom- ing missiles, such as those reported in 1979 and 1980,~7 could have catastrophic results for the entire planet. Finally, despite their apparent sophistication, new technologies may be inadequate for many tasks demanded of them. Computer- based automatic target-recognition systems may be able to identify tanks on a battlefield, but inadequate to distinguish between friendly and hostile tanks. Businesses that now rely on computers for the performance of critical tasks may still be frustrated by their inability to adapt their computers readily to a changing business environment. New computing technologies also raise the issue of their cost. New technology is generally expensive, and thus it can benefit only those who can afford to acquire it. For example, high-resolution monitors installed in schools and connected to a national network may enable students to view images stored in national archives, but schools that can barely afford basic school supplies will not be the first to acquire such monitors. Useful electronic information in the form of software, data files, and database access is often sold as a high-priced specialty item rather than a high-volume commodity for consumption by all. Without public policy and moral commitment to notions of universal access, the information revolution may increase the gap between the haves and have-nots to our collective detriment. It is unlikely that society will be willing to give up the benefits that computing confers upon it, but society is rightly concerned with the problems that computing can cause or exacerbate. This concern generates opportunities for computer scientists and engineers to in- vestigate the development of even newer technologies that deliver more of the benefits but with fewer of the attendant costs. THE RELATIONSHIP BETWEEN THE FEDERAL GOVERNMENT AND CS&E RESEARCH The history of CS&E in both academia and industry reflects the strong influence of strategic investments by federal agencies. These investments have funded work of direct and immediate relevance to government responsibilities (e.g., the use of the first electronic com
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 29 puters for military purposes) as well as work dispersed in large part through the academic research community, the latter especially so in more recent years. Without these investments, the computer indus- try and indeed the information revolution would have taken off much more slowly. While the nature and allocation of federal investment in CS&E have changed over the past four decades, the substantial rise in constant dollars of federal obligations in the last 15 years suggests that the development of advanced computing capabilities through the support of CS&E research is increasingly understood by the federal government to be essential to the missions of many gov- ernment agencies as well as to the welfare of the nation. A variety of federal agencies support research in CS&E, includ- ing the National Science Foundation (NSF) and a few mission-orient- ed agencies, e.g., the Department of Defense, the National Aeronau- tics and Space Administration, and the Department of Energy. Mission-oriented agencies support basic and applied research with the potential to contribute to their missions, while the NSF supports less directed research. These four agencies accounted for 92 percent of CS&E research in FY 1991, both basic and applied, as indicated in Table 1.1. The High Performance Computing and Communications Program, discussed below, promises to have a substantial impact on CS&E research in the next several years, since it calls for substantial interagency cooperation and substantial funding increases to support high-performance computing and communications. Figure 1.2 illustrates that support for CS&E research (basic and applied taken together) to all performers has increased substantially in the last decade, as would be expected for a new and intellectual- ly growing field; funding for academic CS&E research exhibits a sim- ilar trend. (Some readers may object to the grouping together of basic and applied research. This has been done for reasons that will become apparent in Chapter 2, but the general trend also holds for basic research alone.) The federal government is also a prodigious consumer of infor- mation technology and related services, budgeting some $24 billion for information technology in FY 1992.~9 Such expenditures reflect a much broader interest in computer technology than might be implied by the government's research investments alone. Government com- puter use cuts across agencies and sometimes stimulates develop- ment by the private sector of new technologies to meet government needs. The two agencies that account for the largest fraction of federal obligations for academic CS&E research are the Department of De- fense and the NSF. (A more extended discussion of federal agencies
30 COMPUTING THE FUTURE TABLE 1.1 Federal Funding (in FY 1991 dollars) for CS&E Research and All Science and Engineering (S/E) Research, FY 1991 Agency Computer Science Research ($ millions) Cumulative Percentage of Total for Computer Science All S/E Research ($ millions) Defense National Science Foundation National Aeronautics and Space Administration Energy Commerce Interior Environmental Protection Agency Transportation Agency for International Development Treasury Health and Human Services Agriculture Education Housing and Urban Development Federal Communications Commission Other Agenciesa TOTAL 418.7 122.7 52.2 33.3 18.4 11.4 8.3 6.1 3.6 1.7 1.5 1.5 0.9 0.2 0.1 680.6 62 80 87 92 95 96 98 99 99 99 00 00 00 00 00 3,805 1,847 3,463 2,963 444 549 343 146 290 22 8,201 1,177 157 11 631 24,051 NOTE: Table reflects the final disposition of federal obligations for FY 1991, includ- ing congressional action and administration budget reprogrammings in response to congressional action. Figures for "computer science" are assumed to include com- puter engineering. aOther agencies that supported some type of basic or applied research, but not in computer science, include the Arms Control and Disarmament Agency; the Tennessee Valley Authority; the Departments of Labor, Justice, Veterans Affairs, and State; the Smithsonian Institution; the Nuclear Regulatory Commission; and the International Trade Commission. SOURCE: Data from Division of Science Resource Surveys, National Science Foun- dation. supporting CS&E research is contained in Chapter 7.) Among feder- al agencies, the Department of Defense is the largest single funder of CS&E research; in dollar terms, it also accounts for the largest single share of academic research (Figure 1.3~. Defense Department sup- port for CS&E research has contributed directly to many areas that have had a profound impact on computing practice today: time
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 800 700 en 600 o cat 500 400 - u, 300 o . _ 200. 100 o To all rec p ~. ~ / / To academia, 0~_~_~_~' 1977 1979 1981 1983 1985 1987 1989 1991 Fiscal Year 31 FIGURE 1.2 Total federal 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 Federal Fundsfor Research and Development (Federal Obligations for Research by Agency and De- tailed Field of Science/Engineering: Fiscal Years 1969-1990), Division 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, Na- tional Science Four~dation. Basic data (in then-year dollars) for academia taken from Federal Funds for Research and Development (Federal Obligations for Research to Universities and Colleges by Agency and Detailed Field of Science/ Engineering: Fiscal Years 1969-1990), Division of Science Resource Studies, National Science Foundation. Figures include both "computer science" and "mathematics and computer science, not elsewhere classified." Constant dollars calculated from GNP deflators used in National Science Foundation, Science and Engineering Indicators, 1991, NSF, Washington, D.C., 1991, Table 4-1. sharing, networks, artificial intelligence, advanced computer archi- tectures, and graphics. Of course, it is not surprising that a mission- oriented agency would tend to favor research focused on developing operational prototypes; what is striking is that these research projects, initially justified on the grounds of military utility, have yielded such a rich harvest of civilian application.
32 500 400 En - o ~ 300 CM o COMPUTING THE FUTURE To all recipients ~ , 1 .- To academia Use ~-~ f =~^ 1 1 1 1 1989 1 991 1977 1979 1981 1983 1985 1987 Fiscal Year FIGURE 1.3 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 and academia were taken from the corresponding sources cited in the caption for Figure 1.2. The NSF is the primary supporter of academic research in CS&E, as measured by the number of individual investigators supported. It also contributes the second largest share of federal obligations to CS&E research, and almost all of that support goes to academia. Fig- ure 1.4 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 disciplines start at much higher levels. The NSF is the primary federal supporter of investigator-initiated CS&E research. Research agendas within the research directorates of NSF tend to reflect the needs and interests of the field as a whole, although program officials do exercise judgment in determining the appropriate mix of research topics being investigated. Moreover, since the mission of the NSF is largely to support basic research- which the U.S. government defines as research without application in mind (more on this point in Chapter 2)-research supported by NSF is likely to be farther removed from commercial or applications
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 33 oriented impact than research, such as that supported by much of the Defense Department, that is aimed specifically at developing opera- tional prototypes or demonstrating concept feasibility. That said, the NSF has supported research in CS&E that has had a substantial impact on computing practice. For example, ire the 1960s the NSF supported the development of BASIC, a computer language designed for ease of learning that is used in some applications even today. Programming environments (i.e., systems used to support groups of programmers working together in constructing, testing, and maintaining programs) and many important software packages for numerical analysis (e.g., LINPACK for linear algebra) have bene- fited from more recent NSF support. NSF-spor~sored work on image processing in the 1970s and 1980s has led to better imaging scanners . . . in mec .lclne. Finally, both NSF and the Defense Advanced Research Projects Agency (DARPA) of the Department of Defense supported substan 150 140 130 120 en 110 o CM 100 90 80 70 60 50 40 30 ~-~-~-o-I- 20 _ 10 _ O 1 , 1 , 1 1977 1979 1981 To all recipients /~ // I// To academia ] 1 , 1 , 1 , 1 , 1983 1985 1987 1989 1991 Fiscal Year FIGURE 1.4 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 1.2.
34 COMPUTING THE FUTURE fiat efforts in the 1970s and 1980s to build equipment infrastructure in universities for the support of experimental research projects in CS&E. Although federal support for CS&E research has more than dou- bled in the last 15 years, and support for academic CS&E research has about tripled, the number of active researchers in the field has also grown considerably. This has been particularly true in the aca- demic CS&E community, for which the available funding per active researcher has dropped slightly in the last 15 years and more sub- stantially in the last year for which data are available (Figure 1.5~. This drop in available funding per active researcher is consistent with the fact that since 1987, the number of awards made by NSF for CS&E research has lagged behind the number of proposals submit- ted, resulting in a declining success rate for most of this period (see Figure 7.3 in Chapter 7~. Thus the level and adequacy of federal funding for CS&E continue to be a source of major concern to aca- demic computer scientists and engineers. In the last year, a program that cuts across the entire federal government was begun that is expected to have a major impact on federal funding for CS&E. The High Performance Computing and Communications (HPCC) Program began in FY 1992 and is based on a 1989 report by the White House Office of Science and Technology Policy (OSTP)20 that called for a program coordinated across all agencies with responsibilities for or an interest in high-performance comput- ing. The program grew out of efforts by several federal agencies operating under the Federal Coordinating Council for Science, Engi- neering, and Technology (FCCSET) umbrella and in conjunction with the OSTP;2i at present, the initiative involves DARPA, the National Aeronautics and Space Administration, the Department of Energy, and the National Science Foundation, with the participation of the National Institute of Standards and Technology, the National Ocean- ic and Atmospheric Administration (NOAA), the Environmental Pro- tection Agency (EPA), and the National Institutes of Health. The HPCC Program addresses four areas of interest: · High-performance computing systems that will improve the speed of computing by two to three orders of magnitude; · Advanced software technology and algorithms that focus on soft- ware support for addressing certain grand challenges in science and engineering to best exploit high-performance computer systems and tools for more effective development of software systems; · Networking that will support research, development and de- ployment for a gigabit National Research and Education Network
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 100 90 <,, 80 o L) CM 70 60 50 40 30 20 10 o it, \ , C , 1,! 35 1977 1979 1981 1983 1985 1987 1989 Fiscal Year FIGURE 1.5 Federal funding for academic computer science research (in constant FY 1992 dollars) per academic researcher for FY 1977 through FY 1989. SOURCE: Data for federal funding were taken from the sources cited in the caption for Figure 1.2. Data on the number of academic researchers were taken from Table 8.13 in this report. While funding for FY 1990 and FY 1991 has risen (as depicted in Figure 1.2), no definitive data are available on the number of academic researchers working in CS&E for these years, al- though the Taulbee survey of 1990-1991 (David Cries and Dorothy Marsh, "The 1990-1991 Taulbee Survey," Computing Research News, Volume 4~1), Jan- uary 1992, pp 8 95.) reports that the number of CS&E faculty at Ph.D.-grant- ing institutions (i.e., major research institutions) may be leveling off. These considerations suggest that the funding per researcher may have risen in these years. (NREN) and eventual transition of this network to commercial ser- vice; and · Human resources and basic research efforts will focus on ex- panding basic research in all areas of computer science and technolo- gy relevant to high-performance computing and increasing the base of skilled personnel.22 At this writing, the HPCC Program has strong presidential sup- port, and the Congress has authorized most parts of the HPCC Pro
36 COMPUTING THE FUTURE gram for five years.23 However, money for the program must be appropriated yearly, and those portions whose authorization has ex- pired must be reauthorized.24 If the program is fully funded, it will amount to some $1.9 billion over five years in "new Morley," i.e., money above and beyond amounts spent in the baseline budget of FY 1991 (Table 1.2~.25 The amount requested for appropriations in FY 1993 for high- performance computing and communications is an increase of $148 million over the comparable amount in the FY 1992 budget. Table 1.3 describes the funding history of the HPCC Program to date. The magnitude of the requested increase for FY 1993, as well as the fact that the HPCC Program for FY 1992 was actually funded at an over- all level higher than that proposed by the administration, is a clear recognition of the importance of high-performance computing and communications to national goals.26 However, given the central role that NSF plays in supporting the academic CS&E community, con- siderable concern within this community has been raised regarding the fact that the NSF portion of the HPCC Program was funded be- low the requested level. Overall, future federal funding trends are uncertairr. While most federal policy makers appear to understand that CS&E is a field with major impact on the nation's economic health and social well-being, the federal budget will come under increasing stress in years ahead as the result of expected growth in federal budget deficits in the future and the elimination of the so-called peace dividend.27 Also, major budget initiatives require long gestation periods; a significant initiative proposed after August in year N. even if approved at all TABLE 1.2 Proposed Five-Year Funding Profile for the HPCC Program Fiscal Year and Amount ($ millions) Program Area 1992 1993 1994 1995 1996 High-performance computing systems 55 91 141 179 216 Software and algorithms 51 90 137 172 212 Networking 30 50 95 105 110 Basic research and human resources 15 25 38 46 59 TOTAL 151 256 411 502 597 SOURCE: Office of Science and Technology Policy, The Federal High Performance Computing Program, Executive Office of the President, Washington, D.C., 1989, p. 46.
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 37 TABLE 1.3 Funding History and Proposed Funding ($ millions) for the HPCC Program, FY 1991 to FY 1993 FY 1992 FY 1993 Requested Agency FY 1991a Reqb ActualC HPCSASTANREN BRHRTotal DARPA183.0232.2232.2119.5 49.7 43.6 62.2 275.0 DOE65.093.092.310.9 69.2 14.0 15.0 109.1 NASA54.072.471.214.1 61.4 9.8 3.8 89.1 NSF169.0213.0200.928.6 125.6 45.1 62.6 261.9 NIST2.12.92.11.1 1.0 2.0 0.0 4.1 NOAA1.42.59.80.0 10.4 0.4 0.0 10.8 EPA1.45.25.00.0 6.1 0.4 1.5 8.0 NILE13.517.141.34.2 22.6 7.2 10.9 44.9 TOTAL489.4638.3654.8178.4 346.0 122.5 156.0 802.9 (Percentage of Total) (22) (43) (15) (20) (100) aBaseline FY 1991 budget for high-performance computing and communications. bAmounts requested by the administration for FY 1992. CActual amounts available for obligation (after final congressional action). dAmounts requested by the administration for FY 1993. eHPCS, high-performance computing systems; ASTA, advanced software technology and algorithms; NREN, National Research and Education Network; BRHR, basic re- search and human resources. fFor FY 1992, the administration proposed that the National Library of Medicine (NLM) participate in the HPCC Program. In FY 1993, the administration is proposing that the entire National Institutes of Health complex (of which the NLM is a member) participate in the program. SOURCES: Office of Science and Technology Policy, Grand Challenges: High Perfor- mance Computing and Communications, The FY 1992 U.S. Research and Development Program, p. 24; Office of Science and Technology Policy, Grand Challenges 1993: High Performance Computing and Communications, The PY 1993 U.S. Research and Develop- ment Program, p. 28 steps in the process, is not likely to appear in the budget until year N+ 3.28 A key feature of the HPCC Program is that it is framed within the context of specific applications of computing the so-called grand challenges. These grand challenges are "fundamental problemis] in science and engineering, with potentially broad economic, political, and/or scientific impact, that could be advanced by applying high I I t=~U"1~. The HPCC Program recogniz- es that major improvements in computing performance relevant to these grand challenges will be possible only through high-level col- laboration among computer scientists and engineers and scientists and engineers in the relevant areas. One result has been that mission ~ ~ ~ ~. ;~ ~ ~- ~29
38 COMPUTING THE FUTURE agencies not typically associated with CS&E research (e.g., NOAA, EPA) are assigned significant roles in the HPCC Program. THE RELATIONSHIP BETWEEN CS&E AND THE COMPUTER INDUSTRY One primary reason underlying the remarkable successes of CS&E is the extraordinarily fruitful interaction between academia and in- dustry. Academia has supplied the computer industry with many CS&E graduates at all levels well-grounded in the best that academic CS&E has to offer. Furthermore, the interchange of research ideas and problems in both hardware and software has been strong and plentiful. For example, universities have contributed greatly to new computer architectures such as the Hypercube, the Connection Ma- chine, and reduced-instruction-set computing. Workstations are an- other example of fruitful collaboration between the majority of com- puter manufacturers and the leading CS&E research universities (Box 1.7~. In addition, many ideas on specific applications developed in the academic environment, including document-preparation systems, com- puter graphics, and database systems, resulted in a large number of start-up companies that have had a significant impact on society at large and also on well-established computer and software manufacturers. Industry has been a fountain of creative intellectual ferment for academia as well. Industry investment in CS&E research is consider- able, estimated by the committee to be in order of magnitude compa- rable to the federal expenditures on computer science research.30 These investments have resulted in many computing innovations (Table 1.4) and have spun off considerable academic research. For example, the first compilers for translating high-level programming languages into machine language were invented in industry; this ground-breaking work launched hundreds of subsequent university research projects. The Unix operating system and the C programming language origi- nated at Bell Laboratories; however, Berkeley computer scientists have extended and modified the original concepts into a new version of Unix that now enjoys wide acceptance as the basis for highly interop- erable and portable open software systems. In short, ideas first na- scent in the industrial sector have often given impetus to academic research in CS&E. Academic research and industrial CS&E research have both ad- vanced the computing state of the art, but the differences in perspec- tive between academic and industrial researchers are substantial. These differences are important for academic researchers to understand. Companies in the computer industry employ researchers to give
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 39 them a competitive edge in brixlgirlg new products to the market- place or in improving existing products. Industrial researchers influ- ence products through the creation of innovative ideas, but to be useful to product developers, these ideas must be taken to the point of product viability. At the same time, the only viable mechanism for the continual replenishment of the intellectual capital of these researchers is for them to be active cor~tributors to the research enter- prise. Academic researchers have different goals. They may seek knowledge for its own sake, circumscribed in part by the availability of funding, but are not necessarily bound by the need to translate TABLE 1.4 Computing Innovations to Which Industry Has Contributed Innovation Companies Contributing to Related CS&E Research Fortran Unix operating system, C programming language Workstations Microprocessors Supercomputers Minicomputers Local area networks Reduced-instruction-set computing (RISCs) Relational databases IBM AT&T Bell Laboratories Xerox Palo Alto Research Center Intel Cray Research Digital Equipment Corporation Xerox; IBM; Bolt, Beranek, and Newman IBM IBM
40 COMPUTING THE FUTURE TABLE 1.5 Problem and Project Characteristics Tending to Favor Academia or Industry Academia Industry Work directed primarily at the creation of new knowledge Small projects Work with potential benefit in the long term Systematic investigation, theory-building Work directed primarily at improved products and greater competitiveness Large projects Work with potential benefit in the short and medium term Interdisciplinary innovation new ideas into viable products. Table 1.5 depicts some of the charac- teristics of work in CS&E that tend to make a given problem a better fit to the environments offered by either academia or industry. Research can thus be viewed as providing a "service function" to those who develop the nation's computing capability, i.e., the prod- uct developers. In some cases, research provides technologies that are directly applicable to products' even though the need for those technologies may not be known by the product developers. But in many other cases, research is needed to systematize the sometimes ad hoc discoveries and inventions that arise from the practical imper- ative. Entire new research areas in CS&E have developed in this way: operating systems (from the need to use advanced computer systems), database technology (from the need to manage large vol- umes of data), computer security and encryption (from the need to ensure privacy), image processing (from the need to handle pictorial information), and data compression (from the need to reduce trans- mission times of data objects). In all these cases, a rudimentary prac- tice preceded the formulation of the subfield. The role of systematic research was then to systematize, generalize, and clarify the con- cepts, so that systems of much greater capability could be built more easily. This is another way of expressing the important concept that theory often lags practice. If researchers are to perform research that is relevant to products, they must understand the objectives and capabilities of the product developers of industry. To the extent that they can anticipate the future needs of product developers, researchers can address the con- cerns of product development managers by carrying their innovative ideas far enough to demonstrate both their value for products and the improbability of nasty implementation surprises. Box 1.8 de
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 41 scribes some of the things that researchers need to understand about the development environment. How is research best coupled to practical ends? The literature on technology transfer is vast, and a comprehensive examination of tech- nology transfer issues is far beyond the scope of this report; Box 1.9 describes some of the issues posed by technology transfer for the computer industry. Still, in its examination of technology transfer in the context of academic CS&E, the committee identified one point as particularly ~1111111 .. ~ , ~ , .1~111~111 ;; ..;;;;;;;;;'"~.'2'U"2"'" "' " ';;'' " ' ' ;::':' ; ' ' ';'' ' ' ';' ' ;;" ' ' ';''' ' ' ' " "' "' ;"':'' '' "" "' ' ' '' '' ' ' ;'" ;. ;.'"''''''m'''"~"2"''''"""""''' '"' ''''''"' ;'''''''';''';'';'"' '';;;;;;;;;;;;""n'''''o'" ' " ;;' ' :; ' ' ' ' ' ';' '' '';' ' "' ';;' '' ' ''' ' ;'' :'':' ' ' ' ' '' ' ' ' ""':' ' ' ' '' ' ' ' ' " ' ' ;'2'o'2''n''''2"2''' '''' ''''''";' '' ';' ''';'"';';;'''''" ';;'' '"';'';;' ;;;;;;;; ;;;;;;; ;;''2y'''e' ' '; ' ' ' ;' ':::' ' ' "' '' ' ' '' ' ' " ' '"' ";;; ' ' ' ' ' ' ' ' ': ' ' ' '' ' ';' ' ' '"'" '"' ' ' ' ' '' ' ;~"'a""'"~""""''''''' ' " ' ' ' ' 'i' ;'';;;;"' ' ' ' ' I;, ''' ' ';;' ' ' '' ';;' ; ' ' "'' :;;;' ' ' ' ' ' ;;;;';;; ;;;;;;;;;;;;;;'''''a'''n'""""'';; ;' ;' ;" ' ' ';"' "' ' ' ;'' '';''' ' '"' ' " '' : ' "''' ' ' ' ' ' ' ' ';'' ' ' '" '' ' ' ' ' '' ; ''''2m'''e''r''2""" " " ':' ' ';" ;: ' ;'; ;'' ;; ;;; ;; ;;;;;;;; ;~;K ; ;; ; ; ;;; ;;; ; ; ; 2'"a'"'2'p'''''p""2'""2" ' '' ' ' ' '";' ' " ' ' ' '"; ' ' '; '' ' ';' ;' ' ;'';;;" "" ;" ;' ' ';' ~1~1-- ~ P)) ~ ~1~- · . T. . ,.~1~1~ 1~ ;;; '';;;;;;'"'''p'""'~''"'""''" ';;;; ' " ' ' ' ;" ' ;' ' ;';' ' '' "'' "a;' ;' ' " ;" '' ' 'as" ' ' ' ; "''"'~,',''2""m'"~"'"'""''"''' '"' '' '";''' ' ''i' " ;'' ' " ' ''' ;;;;;;;;;;;;;tO """; ; ; ; ;;; ;; ;; ;; ;; ;;; ; ; ;; ; ; ;'2 ' ~ 5' ' ; ; ;; ; ;; ; ;;;; ;;;; ; ; .;;;;;;;;;;;;;''"2c2''''';' '';;' '' ";' '';' ;' '' '" ' ' "';' ' ';; ' ' ' ;;" ';;; '' ' '' " ~ ' ' '' ' ' '' ' ' '" ' ' ' ' ' ' ' ' ' '' ; ;; '""p't."'"'"' " ''' " ' '' '' '"" ' '''' ''"' ' ';' ' ' ' ;';;' " ' '' ' ; ;;' ;;;;'' '';;;' ';' ' ';;;' ;' ' ';';;' '' ';;;; ; .;; ;;;;;;;~"''2e'"2a''' ' ' ' ';' ' " ';' ;' ' ';' ;' "' '''' ';"' ' ;' ''' ' ' ' ' " ' ' ' ' ' ' ' ' ' '' ' ' ' ' ' '' Il~__S=y~. ~ ~! clam the fact that newt idea may Hive ~ substant~t Impact Qd~ the I ~1 ~tend- 1~ ~ ~1
42 COOPING ME FUT-E ~III!I ~=II~C 11 1 ~illi~#lll~l~I>~ 66-IlllllIl~=lill~lElllS./lllll3Sllll3~2I}11113JI-11 (~#~l~II(I~lIII I~llIl~ 1 ~s~sS:~sS:~s~s~s~s s~sss~s~s~isiss~s~s~s~s~sss~s~s~<s<~s~s~s~s~sssssis~ssssiss~s~sss~s~s~s~s~s ssi~s~ssissss~sss~sss~s~s~sss~s~sss~s~s~s~s~..ssssssss~sssssssss~.,. s~s~s~ssss~ss~s~sssss~sss~s~sssss~s~sss~s~sss~s~s critical the routes by which research knowledge can be transferred to hose who can benefit hom iL SpecHica~y, there ~ a broad consensus that knowledge ~ best transferred by moving or using the people who understand the new technology, believe ~ it, and are motivated to fix and solve any problems that may arise during the transfer>] The presence of such individuals is also reassuring to decision makers because the former are thor- oughly familiar with the new ideas. Technology transfer via publica- hons (ne~sIetters, papers, technical reports" documentation, man- agement decree, or one-time Workshops is far less effective. One reason that people are more effective conduits than paper in this context ~ that technical concepts have to be adapted for product usage, often substantinUy, in the give-and-take essential between con- ception and use. Individuals famiLar with the new technical concept
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 43 are generally in the best position to perform dynamic adaptation, while publications and paper are static. Against this background, the committee notes that many current industrial affiliates programs university programs designed to fa- cilitate greater contacts between industry and university research- emphasize prepublication access to research results, but not frequent interaction. The principle that people are more important than paper seems to govern the most successful mechanisms for technology transfer: · Institutional collaborations between academia and industry that provide for sustained, side-by-side contact between the participants. In the absence of such intimate contact, academic research can go off in directions that are essentially irrelevant to commercialization. · CS&E faculty and graduate students who form start-up com- panies. Small start-up firms have been responsible for a dispropor- tionately large share of new commercial applications, often exploit- ing research and good ideas from elsewhere,32 although work funded today by venture capital tends to have a short-range focus, may be less ambitious or risky than research without clear commercial appli- cations, and is increasingly scarce.33 · Graduating students who bring knowledge of the most recent developments in CS&E to the companies that hire them. Of course, if such students are inadequately exposed to these developments, they cannot perform this function; thus undergraduate and graduate edu- cation that reflects the best the field has to offer is a sine qua non if this route is to be effective. If such cooperative efforts are to succeed, it is important that industry understand important values of academia, and vice versa. Thus successful cooperative efforts will tend to involve minimal re- strictions on academic publication, e.g., delays in publication of at most a few months. In practice, such limitations may have little practical significance either to academics34 (since preparing a paper for publication often lags the obtaining of research results by several months to a year or more in any event, and the appearance of a paper in print often takes another six months or more) or to industry (since the duration of the advantage that industry reaps from product inno- vations is often measured in months rather than years). Further- more, papers often emphasize concepts and techniques, which are usually not as sensitive as proprietary details that tend to be relative- ly uninteresting from a scientific perspective in any event. In return, academia must understand industrial concerns. One such concern involves intellectual property rights, a new and evolv- ing area of both legal and ethical concern. In addition to the issues
44 COMPUTING THE FUTURE , ~ _...... I ·~i~¢ ·!-~- ~ :::;:: ::: ::: ::a;: : :::: :::::: :::::: :: ::: ::-: :::::::: :::::::::::::::::::: :: ::.::::: : ; : :: ::::::; :: :::::: ::::::::::::.. ::: :: : :::~::::: ::: :::: : ::::::::::: ::-::: : ::: : ::::::::::::::: ::::::~:~::::i :: :: ::::::::::: :: :::::::: :: ::~:::::::: ::::::: ::::::: ::::::::::::::::::::::: ~1~-.~.,t,~ ' · 11 : :: :::::::: :::...::::: : ::::::: : ' 1 ' ' ' '' "' ' ' ' ':'' ' ' ' ' ' ';' ' ' ' ' ' ' ' ' ' ' ' " '" ' ' ' ' ' ' ' . ... .. . ~ . ~ a. . ~ . . ~ =~lega~l~llll :::::::::.:. ::: ::-:i :::::: ::::, ::::: :::: ::::: ::: ~1~1~1~ Marl IET~1~3i={~ss~o~1~1~1i~ Bee 1~ Except an l t '''"'m'22a""e'S.""' ~ 1991. raised by the intangibility of software (Box 1.10), other concerns arise with respect to the knowledge gained through or with that software, the conflict between possible patent or copyright rights of the people who wrote that software versus the people who financed it versus the people who use it, and the granting of recognition to those who have done the work while protecting the entrepreneurial rights of the sponsors. As one example, many companies in the computer indus- try cross-license their patents with one another to ease the process of bringing individual products to market. But in an environment of pressures for exclusive licenses to maximize revenues (pressures of- ten exerted by universities) and legal uncertainties regarding patent and copyright protection for software, the interests of academia and industry may diverge. However these issues are resolved in any given case, resolution takes time. Universities and companies that
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 45 have made use of umbrella agreements covering all joint work have found that the time between initial contact and final settlement of terms has been sharply reduced. THE CHANGING ENVIRONMENT FOR ACADEMIC CS&E In its infancy and adolescence, academic CS&E has experienced rapid growth and progress. Support for the field increased at a fast clip, and the founders of the new field sustained a high degree of productivity. Computers themselves, once housed in a few large buildings, are now everywhere on office desks. But many important changes are pending in the intellectual, eco- nomic, and social milieu in which academic CS&E is embedded. Per- haps the most important is that a solid record of success increases expectations of those inside the field for continued support and out- side the field for continued practical benefits. Fiscal constraints faced by the major funders of research in this country the federal govern- ment and industry are likely to result in greater pressure to trim research budgets and at the same time generate increased pressure for research to produce tangible benefits. Against this backdrop, several additional changes must also be considered. Changes in the Computer Industry The computer industry is itself undergoing massive change. The influence of formerly strong players such as Data General, Unisys (and its presecessors), and Control Data has waned considerably in the past 20 years, and an environment in which IBM and Apple Com- puter are motivated to collaborate is a different one indeed. Interna- tional competition is on the rise. And, although today's computer industry was built primarily on the sales of large mainframe comput- ers to a relatively few institutions, the computing environment of the future will emphasize to a much greater extent computers as con- sumer-oriented items tools for the masses. In this environment, computing technology both hardware and software will be specialized for intellectual work in much the same way that electric motors are specialized for physical work it will be invisible but ubiquitous. Just as electric motors are an important but invisible part of heaters, washing machines, refrigerators, and alarm clocks, so~also computers and software are today or will be embed- ded in telephones, televisions, automobiles, and lawnmowers. (Ac- tually, it will not be surprising to find them in washing machines and
46 COMPUTING THE FUTURE refrigerators as well.) It is the increasing ubiquity of computing that has led many analysts to predict the eventual convergence of com- puter, communications, and entertainment technology and the emer- gence of information appliances that are dedicated to specific tasks (such as pocket calendars or remote library access devices). New computer systems will be increasingly portable and are likely to be interconnected to each other or to information service providers, and they may well embody new computing styles such as pen-based com- puting. Accompanying these changes in the computer industry per se are other major changes that are affecting all industries. In particular, changes in the business environment portend vastly greater global- ization and time compression. To survive, let alone prosper, indus- try in the future will have to respond to a much larger range of competitors than in the past, and to respond much more rapidly than it has in the past. For the computer industry, these changes mean that products will have to be fitted to customer needs much more precisely. Since customers are interested in computing technology primarily for its value in solving particular problems, knowledge of the customer's application will become more and more important; such knowledge will most likely become embedded in software written to serve these applications. Since customers will be understandably reluctant to abandon substantial investments in hardware, software, and human expertise, it will be necessary to design new products with a high degree of compatibility with earlier generations. Indeed, even today many customers are unable to keep up with, let alone exploit to best advantage, the capabilities of new computing technologies. Since the particular computer products needed by customers cannot be antici- pated years in advance, industry will have to place greater emphasis on reducing the time to market for new products; thus tools, technol- ogy, and approaches to design (e.g., rapid prototyping) to facilitate shorter response times will be necessary, especially for the now la- bor-intensive software sector. Finally, greater concerns about competitiveness will increase fi- nancial pressures on the computer industry, just as they affect all other industry. In an environment of cost cutting, activities that can- not demonstrate an impact on the bottom line will be highly suspect and subject to reduction or elimination. Thus it would not be sur- prising to see industrial research laboratories shift their focus to ef- forts with a more "applied" flavor in the quest of their parent compa- nies for competitive advantage.35 Such a shift may already be starting to occur: the strong connection between CS&E and computing prac
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 47 lice has led to strong demand from the computer industry for indi- viduals who are "system builders," making it more difficult for aca- demia to compete effectively for such individuals. Cost-cutting pres- sures may also be reflected in the willingness of the industry to continue unchanged its practices of donating equipment to academic CS&E departments. These donations (or reduced-price sales) account for a substantial amount of the equipment that these departments use for research and educational purposes.36 Structural Changes in Academic CS&E From a personnel standpoint, the CS&E field has undergone tre- mendous growth in the last decade. For example, according to the Office of Scientific and Engineering Personnel of the National Re- search Council, U.S. Ph.D. production in CS&E grew from its 1979 level of 235 graduates per year to 531 in 1989.37 The number of undergraduate degrees awarded per year grew by more than a factor of three and may be rising again. The number of academic doctoral- level researchers working in CS&E grew from 1052 to 3860 over the same time period, and the number of individuals who have doctor- ates and are teaching CS&E increased from 1613 to 5239.38 The me- dian age of doctoral faculty who teach CS&E grew from 38.4 in 1977 to 43.4 years in 1989, which was about the median age of all doctoral scientists and engineers regardless of field in 1981. (Chapter 8 dis- cusses these and other human resources trends in academic CS&E.) Tremendous growth characterizes the intellectual side of CS&E as well. While it is of course difficult to document in quantitative terms the intellectual maturity of a field, it is nevertheless the judg- ment of the committee that CS&E as an intellectual endeavor has indeed come of age. Although as an organized and independent intellectual discipline it is less than 30 years old, CS&E has estab- lished a unique paradigm of scientific inquiry-a computational par- adigm that is applicable to a wide variety of problems and has be- come the base on which a critical enabling technology of the next century will be built. The opening pages of Chapter 3 and the sec- tion "Selected Accomplishments" in Chapter 5 discuss the accom- plishments and the research paradigm in greater detail. Changes in the University Environment Academic CS&E will be affected by the university environment, an environment that is itself in the midst of remarkable changes.
48 COMPUTING THE FUTURE One major issue is the fact that the compact between the federal government and university research developed in the 1940s and 1950s is under increasing pressure. Implicit in this compact was the under- standing that placing decisions regarding the course of basic research in the hands of the investigating scientist would lead to substantial social and economic benefits as the result of government support for such research.39 However, recent events such as congressional inter- est in alleged abuses in government funding of university research40 suggest that pressures for accountability will increase in the future, and it is entirely plausible that accountability for research will re- quire concrete demonstrations of positive benefit to the nation. Financial considerations also loom large. Universities everywhere are suffering from ever tighter budgets, and it does not appear that these exigencies will abate in the foreseeable future. Apart from the difficulties that all academic disciplines will face in matters such as faculty hiring, academic CS&E departments will face particular prob- lems in maintaining infrastructure to meet the field's research needs. As noted earlier, many research problems in CS&E are driven and motivated by the upper bounds of performance at the cutting edge of computing technology (whether these edges result from so- phisticated new components or novel arrangements of older compo- nents). The availability of state-of-the-art systems to address these problems is therefore critical if CS&E departments are to stay at the cutting edge of research, whether in software or hardware. Howe~r- er, 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 per- formance. Compounding the problem is the fact that a system that is state of the art today may not remain so for very long.4i Large and often recurring replacement costs will be necessary for departments to remain at the hardware state of the art. Capitalization for educational purposes is also an important as- pect of acquisition budgets. CS&E students (especially undergradu- ate students) may not need access to computing equipment that is absolutely at the cutting edge, but all too often undergraduate CS&E students must make do with personal computers that were acquired in the mid-1980s and that often cannot run modern software. When they must use hardware whose capabilities are so limited, students are forced to struggle with machine limitations rather than focusing on central concepts that could be more clearly illustrated with more powerful machines. For the teaching of some topics, hardware that is so limited in performance is not effective as a pedagogical tool.
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES SUMMARY AND CONCLUSIONS 49 Computing has become indispensable to modern life, and every computer in use today is based on concepts and techniques devel- oped by research in CS&E. Future advances in CS&E research will have a similar impact: they will increase the use of computing and the effectiveness of computing. But after several decades of vigor and growth, the CS&E field is facing a very different environment. Academic computer scientists and engineers the primary group ad- dressed in this report will have to cope with a host of new challeng- es, some arising from the remarkable successes of the discipline (e.g., the spread of computing to virtually all walks of life) and others from factors er~tirely outside the discipline (e.g., pressures on federal research support). How should the community respond? As Chapter 2 describes at length, the committee believes that academic CS&E must begin to look outward, embracing rather than eschewing other problem do- mains as presenting rich arid challenging topics for CS&E research. NOTES 1. Business Week, October 28, 1991, p. 120. 2. Written testimony of Jack L. Brock, Information Management and Technology Division of the General Accounting Office, to the Subcommittee on Science, Technolo- gy and Space of the Senate Commerce Committee, March 5, 1991, p. 6. 3. Written testimony of Jack L. Brock to the Subcommittee on Science, Technology and Space of the Senate Commerce Committee, March 5, 1991, p. 4. 4. The Bossiness Week 1000, 1991 Special Issue, pp. 174-175. 5. For example, a floppy disk with a word-processing program on it and one with- out the program on it have identical weights, but the first disk is much more useful and valuable. 6. The Business Week 1000, 1991 Special Issue, p. 167. The "Entertainment" catego- ry lists five major corporations. 7. U.S. Department of Commerce, U.S. Industrial Outlook 1991, U.S. Government Printing Office, Washington, D.C., 1991, p. 39-6. 8. The Business Week 1000, 1991 Special Issue, p. 178. 9. The GNP of the United States was $5465.1 billion in 1990 (U.S. Department of Commerce, Survey of Current Business, Volume 71(7), July 1991, p. 5). For 1990, the Computer and Business Equipment Manufacturers Association (CBEMA) estimated revenues derived from computer equipment at $153.7 billion (p. 26), from computer software at $92.4 billion (p. 24), and from telecommunications equipment at $61.7 billion (p. 26); in total, these categories accounted for about 5.6 percent of the GNP. (Page references are for CBEMA Industry Marketing Statistics Committee, The I~forma- tion Technology Industry Data Book: 1960-2000, Computer and Business Equipment Manufacturers Association, Washington, D.C., 1990.) A different set of estimates is provided by the U.S. Department of Commerce (U.S. Industrial Outlook 1991, U.S. Gov- ernment Printing Office, Washington, D.C., 1991): computers and peripherals, $71 bil
50 COMPUTING THE FUTURE lion (p. 28-1); software, $29 billion (p. 28-15); telephone and telegraph equipment, $18.5 billion (p. 30-1); radio and TV communication equipment, more than $55.8 bil- lion (p. 31-1); electronic information services, $9 billion (p. 27-2); data processing and network services, $31 billion (p. 27-3); and computer professional services, $44 billion (p. 27-4). Taken together, these categories totaled 4.7 percent of the GNP. 10. In this report, the term "computing" denotes both the electronic activity taking place when computers are being used and the problem-solving activities to which computers are directed. "Computing practice" or "the practice of computing" denotes computers used as tools for solving problems in domains not intrinsically related to computers themselves. "Computer science and engineering" (CS&E) is used more narrowly to denote a field whose research and development activities are related to computers per se. 11. The notion of CS&E as a discipline based on theory, abstraction, and design is described in Peter Denning, Douglas E. Comer, David Gries, Michael C. Mulder, Allen Tucker, Joe Turner, and Paul R. Young, "Computing as a Discipline," Communications of the ACM, Volume 32(1), January 1989, pp. 9-23. 12. Personal communication, Donald Knuth, March 10,1992 letter. 13. Frederick Brooks, The Mythical Man-Month, Addison-Wesley, Reading, Mass., 1975, pp. 7-8. 14. Ivan Sutherland, "Computer Displays," Scientific American, June 1970, p. 57. 15. A point-of-sale network is a network of electronically linked cash-register/ter- minals that can capture purchasing information at the moment and place a sale is made (i.e., at the "point of sale") for such purposes as tracking inventory, debiting and crediting funds between customer and store bank accounts through electronic funds transfer, or automatically generating purchase orders for new merchandise. Or, it can perform some combination of these tasks. 16. In 1990, 53 percent of the American public disagreed with the statement that "computers and factory automation will create more jobs than they will eliminate." See National Science Foundation, Science and Engineering Indicators, 1991, NSF, Wash- ington, D.C., 1991, p. 455. 17. Gary Hart and Barry Goldwater, Recent False Alerts from the Nation's Missile Attack Warning System, Report to the Senate Armed Services Committee, U.S. Govern- ment Printing Office, Washington, D.C., October 10, 1980. In 1979, a test tape was mistakenly entered into the missile early warning system of the Strategic Air Com- mand. In 1980, the failure of a computer chip generated two erroneous warnings of . . . .. Incoming missiles. 18. Funding figures have been drawn from various sources of the NSF Division of Science Resources Studies (SRS) series. The NSF SRS Division compiles these figures on the basis of questionnaires completed by the various federal agencies. Thus it identifies what the various agencies believe should be counted under the label "com- puter science." Such self-identification of funds, in the absence of a standard and consistent definition, may easily lead to errors and omissions, especially in the case of projects that contain important CS&E elements but that are not themselves obviously CS&E. For example, the National Institutes of Health does not fund much research that it reports as "computer science" research ($300,000 in FY 1990). Nevertheless, according to an NIH briefing received by the committee, NIH funds some $150 million per year in medical imaging research, research that has a strong CS&E aspect and may even be performed in CS&E departments. Similarly, research funded under electrical engi- neering may be computer design. However, agencies may also label as "computer science" work that may more properly be classified under "applied mathematics."
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 51 SRS figures have been used because they come from a single source that attempts to ensure that trend comparisons can be made. The alternative would have been to dig into the data in detail (i.e., at the individual grant and contract level) for the years in question, an undertaking well beyond the scope of this project. In addition, since individual agencies tend to use the same identification process year in and year out, the SRS figures are likely to reflect trends over time for individual agencies. However, note that when funding figures for FY 1990 and FY 1991 are presented, they are pre- liminary and subject to later revision. Figures presented in the funding charts of this chapter and in Chapter 7 are the sum of line items labeled "computer science" and "mathematics and computer science, not otherwise classified," are in FY 1992 (constant) dollars, and are assumed to include computer engineering. 19. Bob Brewin, "IT Dollars to Inch Up Next Year," Federal Computer Week, April 25, 1992, p. 1. 20. Office of Science and Technology Policy, The Federal High Performance Comput- ing Program, Executive Office of the President, Washington, D.C., September 1989. 21. The Federal Coordinating Council on Science, Engineering, and Technology consists of the heads of all agencies that have responsibilities for issues with signifi- cant scientific or technical aspects. Chartered in the early 1970s and revitalized in 1989 under Science Advisor D. Allan Bromley, its purpose is to provide interagency coordi- nation for activities related to such issues. 22. The 1989 OSTP report articulated specific goals: increasing Ph.D. production in computer science to 1000 per year by 1995, upgrading 25 additional university com- puter science departments to nationally competitive quality, and improving connec- tions between computer science and other disciplines, including the creation of at least ten computational science and engineering departments (p. 40). But neither the legis- lation nor its legislative history mention these specific goals, except to specify that the Congress expects the HPCC Program to be similar to that presented in the 1989 report (Senate Commerce Committee, High-Performance Computing Act of 1991: Report of the Senate Committee on Commerce, Science, and Transportation, Report 102-57, U.S. Govern- ment Printing Office, Washington, D.C., 1991, p. 16). 23. Since the HPCC Program is a multiagency program, authorizations are con- trolled by different committees of the Congress. Five-year authorizations for the NSF, Departments of Energy and Commerce, NASA, and the Environmental Protection Agency were specified by the High-Performance Computing Act of 1991. A one-year authori- zation for the DARPA portion was passed by the National Defense Authorization Act for Fiscal Years 1992 and 1993, and will be revisited in FY 1993. The National Insti- tutes of Health has been operating under the "rolled-over" authorizing legislation of FY 1990 since that year, although a multiyear authorization bill for FY 1993 and be- yond is pending in Congress as this report goes to press. 24. The budget process typically involves four major steps. The first is that the administration proposes a budget, called "the administration's request." The second step is usually that the Congress passes "authorizing" legislation that provides what amounts to an upper bound on the amounts that the Congress may appropriate in later years. Authorizing legislation also generally determines the broad policy out- lines that the administration must follow in implementing the program. Authorizing legislation is often (though not always) based on the broad outlines of the administra- tion's request; for major programs, authorizing legislation nearly always makes some budget or policy changes in the request. In the event that authorizing legislation is not specifically passed for any given fiscal year, Congress often resorts to stop-gap legisla- tion that simply rolls over authorizations from previous years. The third step is that
52 COMPUTING THE FUTURE the Congress passes "appropriating" legislation that provides the administration with the authority to obligate money (i.e., write checks for specific purposes); appropriating legislation is passed yearly. There is no legal requirement that the amounts appropri- ated match the amounts authorized, though in practice amounts appropriated above the authorized figures are rare and amounts appropriated under the authorized figure are somewhat more common. The fourth step is that the administration responds to the congressional appropriation. For example, if the appropriation for the National Science Foundation is lower than that proposed in the president's budget, the admin- istration must decide how to parcel out that cut among the various directorates of the NSF; it has complete freedom to make these decisions, as long as they are consistent with congressional intent on the matter. 25. The difference between the 1991 amount in Table 1.3 ($489 million) and the amount in Table 1.1 for FY 1991 ($680 million) reflects the fact that not all federally funded CS&E research is part of the HPCC Program. Similarly, not all HPCC Program funding is intended for the CS&E community; researchers in other "grand challenge" disciplines will also benefit from the HPCC Program. 26. During congressional debate on the HPCC Program, the administration cited a study that estimated a payback of $10.4 billion in supercomputer revenues from the pursuit (at full funding levels of $1.9 billion over the next five years) of the HPCC Program (p. 119). This study also forecast a cumulative increase in GNP of $172 billion to $502 billion over the next decade (p. 143). See the Gartner Group, High Performance Computing and Communications: Investment in American Competitiveness, Stamford, Con- necticut, March 15,1991. 27. The ending of the Cold War was thought by many to herald an era in which military spending would be sharply curtailed and the savings made available for other purposes. But the budget agreement for FY 1991 between the president and the Con- gress stipulated that military spending and nondefense, discretionary spending would constitute two entirely separate categories and that cuts in one category could not be used to increase spending in another category. This agreement was originally sched- uled to expire in FY 1993, so that the FY 1994 budget will not be subject to this rule. Whether this agreement will continue to remain in effect is not clear as this report goes to press. A very good survey of the pressures on federal funding of the research enterprise is contained.in Office of Technology Assessment, Federally Funded Research: Decisions for a Decade, U.S. Government Printing Office, Washington, D.C., May 1991. 28. David Sanchez, "The Growing, Caring and Feeding of a Budget," NSF Direc- tions Newsletter, STIS DIR-916, Office of Legislative and Public Affairs, National Sci- ence Foundation, Washington, D.C., Volume 4(2), March-April 1991. 29. Office of Science and Technology Policy, The Federal High Performance Comput- ing Program, Executive Office of the President, Washington, D.C., September 8, 1989, p. 8. Some of the grand challenges listed on pp. 49-50 of this document are the predic- tion of weather, climate, and global change; semiconductor design; drug design; the human genome project; and quantum chromodynamics. 30. The committee's estimate is based on an assumption that the half-dozen or so major firms in the computer and communications industry (e.g., AT&T, IBM) employ a few thousand full-time CS&E Ph.D. researchers and hire hundreds of new CS&E Ph.D.s every year (as indicated by the various Taulbee surveys). Assuming that each re- searcher costs an average of $200,000 per year in salary, benefits, and equipment, industrial researchers represent an annual investment of several hundred million dol- lars per year. (This estimate does not take into account the fact that a substantial
COMPUTING SIGNIFICANCE, STATUS, CHALLENGES 53 portion of industrial research is conducted by holders of master's degrees or Ph.D.s from other fields.) This figure can only be estimated due to the fact that reports of corporate R&D spending generally do not disaggregate research and development, let alone research in different fields. However, according to common rules of thumb, research costs tend to be perhaps a tenth of development costs, which are themselves perhaps several percent of gross revenues. Thus the figure of "several hundred million" per year spent on CS&E research is not grossly inconsistent with the $153 billion per year in sales of the computer industry reported by CBEMA in Note 9 above. (One data point on the relative size of research vs. development is that IBM's R&D budget in 1991 was about $6.5 billion, of which 90 percent went to development. See John Markoff, "Abe Peled's Secret Start-Up at IBM," New York Times, December 8, 1991, Section 3, p. 6.) 31. This understanding is echoed in Government-University-Industry Research Round- table/Academy Industry Program, New Alliances and Partnerships in American Science and Engineering, National Academy Press, Washington, D.C., 1986, p. 36. 32. Computer Science and Technology Board, National Research Council, Keeping the U.S. Computer Industry Competitive: Defining the Agenda, National Academy Press, Washington, D.C., 1989, p. 59. The report notes that many successful ideas in software have had their origin in large research investments by big companies and that these ideas have been commercialized by small start-up firms. Though the report refers to research originating in industry, the same is likely true for academic research as well, since the difficulties of commercializing research tend to arise regardless of the re- search's origin. 33. The flow of venture capital to small business had dropped by nearly a factor of two in 1990 compared to its peak in 1987. See "Agenda for Business," U.S. News and World Report, June 3, 1991, p. 62. 34. See also Government-University-Industry Research Roundtable/Academy In- dustry Program, New Alliances and Partnerships in American Science and Engineering, National Academy Press, Washington, D.C., 1986, p. 29. 35. For example, Kumar Patel, a research director at AT&T Bell Laboratories, says that "we have a narrow view of what's important to us in the long run. What we call basic research is what fits the general needs of the company." See "Physics losing the corporate struggle," Nature, Volume 356, March 19, 1992, p. 184. While this article emphasizes shifts at Bell Labs, Bellcore, and IBM away from basic research in physics, the reasons for such shifts are closely related to the business interests of the respective companies. 36. According to an NSF survey, private and industrial sources accounted for about 29 percent of research equipment acquisition budgets for academic CS&E in 1988. See National Science Foundation, Academic Research Equipment in Computer Science, Central Computer Facilities, arid Engineering: 1989, NSF 91-304, NSF, Washington, D.C., January 1991, Table 4, p. 5. 37. Throughout this report, figures related to Ph.D. production are taken from the Office of Scientific and Engineering Personnel (OSEP) of the National Research Coun- cil. As Chapter 8 indicates, these numbers at times differ considerably from figures commonly available to the field, such as those of the Taulbee surveys; these figures also lag the Taulbee survey by a couple of years. However, these figures have been used because the OSEP is also responsible for collecting such data for other fields, making the data usable for comparative purposes. Reasons for the discrepancies in data from the various sources are discussed in Chapter 8. 38. The number of academic CS&E researchers over time is presented in Table 8.13.
54 COMPUTING THE FUTURE The number of those teaching CS&E in these years is taken from data provided by the Office of Scientific and Engineering Personnel of the National Research Council and includes those teaching computer science, computer engineering, and information sci- ences. 39. This compact is best described in Vannevar Bush, Science the Endless Frontier, NSF-90-8, National Science Foundation, Washington, D.C., 1945/1990: "Scientific progress on a broad front results from the free play of free intellects, working on subjects of their own choice, in the manner dictated by their curiosity for exploration of the unknown" (p. 12) and "Support of basic research in the public and private colleges, universities, and research institutes must leave the internal control of policy, person- nel, and the method and scope of the research to the institutions themselves" (p. 33), as well as the text of Note 1 in Chapter 2. 40. Colleen Cordes, "Audits Indicate 14 Universities Improperly Charged Govern- ment for $1.9 to $2.4 Million in Overhead," Chronicle of Higher Education, Volume 38(10), October 30, 1991, pp. A26-A29; Daniel E. Koshland, Jr., "The Overhead Ques- tion," Science, Volume 249, July 6, 1990, pp. 10-13. 41. In one NSF survey conducted in 1985-1986, administrators from computer sci- ence departments regarded research instrumentation and equipment that was more than one year old (on average) as not "state-of-the-art." See National Science Founda- tion, Academic Research Equipment in Selected Science/Engineering Fields: 1982-1983 to 1985-1986, SRS 88-D1, NSF, Washington, D.C., June 1988, Table B-5, p. B-14.
