Click for next page ( 15


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



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 14
G Hardware INTRODUCTION Computer hardware consists of machines (computers) of varying sizes, costs, and performance levels plus what the committee has labeled base technologies, components from which the machines are built. The range of computers can be seen in the following taxonomy: Class Price Example Supercomputer > $5 million Cray models Mainframe > $500,000 IBM 3090/3080 Minicomputer > $50,000 DEC 8600/~800 Workstation ~ $7,000 Sun 3/60 Personal Computer < $7,000 IBM-PC This chapter summarizes relevant developments in five areas: base technologies; supercomputers; novel parallel processors; work- stations, minicomputers, and multiprocessors; and personal comput- ers and microcomputers. The first four areas embody the most significant hardware tech- nology advances. The discussions of supercomputers, novel parallel processors, and workstations, minicomputers, and multiprocessors address different aspects of high-performance computing. The dis- cussion of supercomputers focuses on conventional large and fast ma- chines, while the other discussions deal with alternative approaches 14

OCR for page 14
HARD WARE 15 to high-performance computing. A brief discussion of PCs is provided because those machines present trends of great practical importance. BASE TECHNOLOGIES The major underlying technologies that drive the evolution of computing and information processing systems fall into four cat- egories: semiconductor device and component technologies, inter- connect technologies, mass storage technologies, and network and communication technologies. This section examines trends in the first three categories; trends in network communication technologies are discussed in Chapter 5. Semiconductor Device and Component Technologies Today's computers include and depend for improvement on sev- eral semiconductor components. Advances in those components are driven by trends in semiconductor material and device technologies (see Appendix A). The term semiconductor refers to a key property of the materials from which semiconductor components are made, but they are also called integrated circuits (ICs) in reference to their design. [C components generally provide either of two functions for a computer: processing capability sometimes referred to as logic or memory. Semiconductor logic components, such as microprocessors and numeric/floating point coprocessors, are important applications of integrated circuit technology. They provide the central process- ing functions for a wide range of low-cost, high-volume computer systems, such as PCs and workstations. As integrated circuit tech- nology advances over time, more processing and support functions will be combined within a single component to increase processing performance without a commensurate increase in component cost. Additional information regarding microprocessors can be found in the discussion of PCs below. This discussion focuses on dynamic random access memories (DRAMs), static random access memories (SRAMs), and applica- tion-specific integrated circuits (ASICs). Dynamic Random Access Memory Components . Dynamic random access memory components are a pacing item In integrated circuit technology and its further development. They

OCR for page 14
16 GLOBAL TRENDS IN COMPUTER TECHNOL OGY TABLE 2.1 Projected Availability of DRAM Components Year DRAM 1978 1982 1986 1988 1989 1992 16 kbits 64 kbits 256 kbits 1 Mbits 4 Mbits 16 Mbits make up a significant fraction of the dollar volume of all semicon- ductor manufacturing. DRAMs are used to meet the high-capacity, Tow-cost read/write needs for ciata storage in Tow-cost and high- volume computer systems. The principal strength of DRAM is high throughput with high density; a weakness of it is aging, but this is be- ing corrected with more sophisticated refresh schemes (e.g., on-chip refresh). Table 2.1 projects the availability of DRAM by size, measured in bits of data storage capacity, en cl by year of development (starting with historical data). The genera] trend is for commercially available, cost-effective random access memories (RAMs) to lag first samples by about two years and first announcement in a research meeting by an additional one or two years. The latter lag time has been lengthening somewhat over the experience of the late 1970s, partly as a result of increasingly more complex technology. A significant development will be the extension of trench ca- pacitor technology, which takes advantage of chip depth to increase feature density, or other competing technology, in production be- yond the 4-megabit (Mbit) DRAM to 16 Mbits or even 256 Mbits (although the integration level for the latter may require a new ap- proach). DRAMs are discussed again later in the section on mass data storage, where they are compared to other storage devices. Static Random Access Memory Components Static RAM technology has inherent performance advantages over DRAM technology.) Consequently, it is more crucial than iSRAM has an inherent performance advantage over DRAM because of its static

OCR for page 14
HARD WARE 17 DRAM technology to high-performance machines, although it is also used in Tow-response-time machines. Clock rates of high-performance computer systems are tied to the speed of the fastest SRAMs avail- able at a given density in that time period. SRAMs also address the Tow-response-time market and should evolve to meet the demands of small high-speed memories, high-performance microcomputer mem- ories, and supercomputer main memories as well.2 3 Appli cation-Specific Integrated Circuits Semiconductor components designed for specific appli cations (e.g., automotive engine controllers, communications controllers) are categorized as ASICs. The competitive position of a company of- fering an ASIC design and manufacturing capability typically de- pends on the vendor's manufacturing technology, performance, and the computer-aided design (CAD) tools available to generate a de- sign. CAD tools are one of the most important pacing items in ASIC development.4 Close interaction between system developers and ASIC designers is essential, a factor that could eventually favor vertically integrated companies. Potential Breakthroughs A potential breakthrough that could overtake semiconductor trends involves advances in biomolecular electronics (nanotechnol- ogy), an immature technology that seeks to replace semiconductor device structures, such as gates, switches, and buses, with molecular structures, such as neurons and neural networks. Research in nano- technology has accelerated during the past several years in Japan, the Soviet Union, and the United States. Currently, research work in this domain can be classified as basic research in biomolecular structures, access (straight decode, no sequencing) and because it does not require refresh. These advantages should help assure a place for SHAM in the near-term markets but it will have to move to the fastest device technology to maintain a perfonnance distinction over DRAM. 2 All these tend to demand byte (or greater) widths and multiple-port capacity. 3It is likely that SRAMs will continue to move to a bipolar complementary metal oxide semiconductor, emitter coupled logic, and, perhaps, eventually to gallium arsenide. 4Cell libraries, which contain basic "building blocks" of circuits used in a design, are including more and more ASICs and system-specific cells, and fewer general-purpose cells. New compilers allow dramatic decreases in cell design time and compilation, which may lead to the use of multiple technologies on a single chip.

OCR for page 14
18 GL OBA L TRENDS IN COMP UTER TECHNOL O G Y with modest progress being demonstrated in laboratories. The most aggressive forecasts of the commercial application of this technology within the computer and information processing industry indicate that initial devices and components may emerge in the 2000 to 2010 time frame. Leading Industry Players The semiconductor area is too diverse to allow broad general- ization regarding global competitive leadership. However, specific trends within several categories of semiconductor components and base technologies can be identified using market share projections and assessments of the quality of semiconductor devices and compo- nents being produced. Further, U.S. prospects are weakened by an eroding position in semiconductor manufacturing equipment. With respect to commodity DRAMs and other memory compo- nents up to 256 bits, Japanese companies (Fujitsu, Hitachi, NEC) and Texas Instruments in the United States are the world leaders. In megabit DRAMs, several Japanese companies, IBM, Texas Instru- ments, Micron, and the Philips/Siemens consortium are all producing 1-Mbit DRAMs, and Al have announced 4-Mbit DRAM production for 1989. In the S RAM arena, the technical leaders are major Japanese corporations and a number of small U.S. start-ups (Cypress Semi- conductor, Performance, IDT). U.S. manufacturers often use better designs with inferior technology, and in bipolar SRAMS there are no significant U.S. suppliers. Several U.S. companies are world ASIC leaders because they have developed the best CAD tools. However, some large Japanese companies, such as Mitsubishi and NEC, have been aggressively re- searching CAD, and improvements in standard ceils and gate arrays are expected. There are two major efforts in Europe: European Silicon Structures, which is concentrating on rapid prototyping of BASIC tools (and eventually wiD have a foundry), and the CATHE- DRAL CAD system under development by the Interuniversity Mi- croelectronics Centre at Catholic University/Leuven, funded by the ESPRIT program of the European Economic Community (EEC). In bipolar technology, overall leadership resides in IBM and in Japanese firms. Unlike Japanese companies, IBM focuses on internal use rather than merchant sales. Texas Instruments is also a major bipolar producer.

OCR for page 14
HARD WARE 19 In gallium arsenide (GaAs) technology, a number of elements are important the source of the raw material, the technology for growing crystals, the production of wafers, and the production of circuits. The largest source of raw gallium is the People's Republic of China (PRC). Several Japanese companies have purchasing agree- ments with the PRC for raw gallium. Japan and the United States are co-equals with respect to crystal growth technologies, principally molecular beam epitaxy and metal-organic chemical vapor deposi- tion. Virtually defect-free wafers of 3- and 4-inch diameters, all of roughly the same quality and price, are available from suppliers in Europe, Japan, and the United States. (Thomson CSF and Philips have recently developed a 6-inch GaAs wafer capability, although it is not known whether production has begun.) Interconnect Technologies Interconnection technology is important within devices en c] at the system or architecture level. Device-level (or first- and second- level) interconnects are discussed in Chapter 3 in the section on packaging technology. A critical component of future high-performance multiprocessor machines and other parallel architectures wiD be low-latency, high- density interconnection at the system level (sometimes referred to as third-level interconnect). In the near term, this implies the use of some type of high-bandwidth "switch" or bus technology. A fundamental driver of advances in computing systems over the past 10 to 15 years has been the ability to scale devices to smaller sizes with corresponding increases in device speect and cir- cuit density. Applying general architecture principles for large-scare machines to smaller systems has been possible without requiring dra- matic changes in the underlying architecture. However, it is likely that the physical limits on device scaling will end this evolutionary driving force, and further progress in achieving higher performance and less costly computing systems will require new architectural prin- ciples and new technologies. For individual devices and small-scare functions, interconnections willincreasingly constrain performance (see Chapter 3), and at the system level, interconnections will also dominate performance and impose serious limits on architectural alternatives.

OCR for page 14
20 Potential Breakthroughs GL OBA L TR ENDS IN COMP UTER TE CHNOL O G Y During the next decade, progress is expected for several emerg- ing technologies, such as optical interconnects and high-temperature superconducting interconnects. However, even the most aggressive forecasts for the commercial use of these technologies in computer and information processing systems indicate that these interconnects may not emerge until after 2000. A breakthrough in one or more of these technologies would dramatically improve the cost and perfor- mance of computing systems. I,eading Industry Players Research on system-level interconnects is conducted in Europe, Japan, and the United States. To date most near-term computer systems with single or multiple processors make use of conventional bus technology for system-level interconnect. More exotic forms of system-level interconnect are still far from the marketplace. Japan's Electro-Technical Laboratory (ETL) is developing an optical (system-level) interconnect system called Di- aJog H. which is still in the laboratory. The University of ErIangen and AT&T's Bell Labs are working on an optical switch (based on an optical logic etaJon), and work on an optical cross-bar switch is under way at Stanford University for the Office of Naval Research. Thomson CSF has had a major research program in optical switch- ing technologies for many years and has been experimenting with a prototype optical switch. While offering Tots of potential in terms of bandwidth and greatly reduced or absent crosstalk, these switches are still a Tong way from solving all the technical problems. Mass Data Storage Mass data storage systems wiB be critical to future high-perfor- mance systems because the latter wiD be expected to address prob- lems requiring improved memory capabilities. The principal pa- rameters of merit are read/write, density, access speeds, archiving capability, and storage capacity. Basic technology for data storage is primarily dependent on ma- terials. However, certain semiconductor, optic, electronic, and mag- netic technologies could also be applicable to data storage subsystems for computer systems. This chapter does not address magneto-optical

OCR for page 14
HARD WARE 21 technologies in depth, because they are already available on the mar- ket. Nor does it address the molecular-engineered technologies, the commercial availability of which is likely to fall outside the time period of this study. Major Technology Trends For more than a decade, computer systems have relied primarily on three distinct technologies for data storage: solid-state DRAM chips for internal memory, magnetic disks for large capacity on- line memory, and magnetic tape for archival storage and off-line storage of large databases that are accessed only periodically. In basic form, these three storage technologies have remained remarkably constant, but each has been driven steadily forward in price and performance by an ever increasing areal density of bits on the storage medium. That is, there have been generally predictable gains in the linear dimensions required in each medium (silicon, thin-fiIm-plated disks, or magnetic tape) to store a bit of data; but linear density increases result in geometric improvements in the data stored per unit area. Further, efficiencies in manufacturing technology and read/write mechanisms have held the media cost nearly constant, and access times have improved gradually. Thus there have been geometric improvements in the cost of storing a given amount of data in each technology and the amount of data that can be held on-line, as weD as substantial improvements in the speed of access to such data. Disl~^ and Tape Components. It can be anticipated that magnetic disks and tapes will be augmented or even partially supplanted by new technologies with even Tower price and better performance, just as punched cards and paper tapes were supplanted in the past. Re- search is under way in Europe, Japan, and the United States on several basic materials that are applicable to mass data storage, and these products could be candidates to replace magnetic disk technology within the next decade. The materials are: conducting polymers, piezo-electric materials, controlled impedance substrates, high-coercivity magnetic recording media, liquid crystals, and re- versible optical recording materials. In particular, read-only optical disks are already available for data distribution, and widespread introduction of read/write optical storage products (magneto-optical) is anticipated shortly. Recent

OCR for page 14
22 GLOBAL TRENDS IN COMPUTER TECHNOLOGY ~ 10 D 1 - 1985 1990 1995 2000 Year Magnetic Tape Magnetic Disk Optical Disk FIGURE 2.1 Storage densities for magnetic tapes, magnetic disks, and optical disks. SOURCE: Data are derived and extrapolated from general industry data and trends such as can be found in the Computer Storage Industry Service and the Technical Computer System Industry Service published by Dataquest. / and anticipated trends in are al density are shown in Figure 2.1 for traditional magnetic media and expected optical disks. It can be seen that improvements by a factor of 10 in storage capacity are expected between 1990 and 2000, implying sharply declining costs en cl lower access times. Demand for data and storage capacity will doubtless also con- tinue to escalate, so that the amount of money actually spent on memory products may not fad much with declining cost per unit capacity. It is worth noting that initial optical products have come primarily from Japan, and that the Japanese are investing heavily in product development. Within the broad domain of magnetic disk components, several categories of magnetic disk components can be definer! in terms of cost, performance, and capacity. A current snapshot of the char- acteristics of these categories is shown in Table 2.2. Although the parameters shown in the table will change over time, they will track along the trends outlined earlier. Both floppy and Winchester disks have become commodities in the world market. Solid-State Technology. ' grated circuits is following The density of data storage on DRAM inte- ___= similar trends. In fact, currently, storage capacity of a memory chip is about the same as that of 1 square inch

OCR for page 14
HARD WARE TABLE 2.2 Current Cost and Capacity Characteristics of Magnetic Disk Components - Transfer Capacity Category (Mbite) Cost Rate ($) (kbite/~) Floppy disks 1-40 100-1,000 60-600 Winchester disks 10-350 500-4,000 600-1,200 Removable media 200-600 4,000-20,000 600-1,200 (RM) disks Large RM disks 500-4,000 >20,000 1,200-10,000 SOURCE: Data are derived and extrapolated from general industry data and trends such as can be found in the Computer Storage Industry Service and the Technical Computer System Industry Service published by D ataquest. 23 of magnetic tape (1 Mbit). But while tape and disk products use part of their areal gains to shrink media size, memory chips actually are growing somewhat in size and thus improving even faster in unit cost. As density improves, so does performance. This can be seen in Figure 2.2, which shows the trends in price per megabyte of storage capacity for solid-state memory and magnetic disks. The cost differential goes from a ratio of about 50:1 in 1985 to a projected ratio of only 2:1 in the year 2000. Since there is no reason to believe that solid-state memory will lose its speed advantages, there is likely to be a major shift toward replacing rotating media with very large scale integrated (VEST) memory, often referred to as "so~id-state disks." Indeed, it is possible that traditional magnetic media will find fewer applications (at least in new products) because of the more favorable price and superior performance provided by optical and wafer-scare semicon- ductor products. Because the Japanese VEST vendors have achieved substantial dominance in commodity memory chips (although they in turn may be squeezed out by Tower cost producers in other parts of the Far East), it is likely that an ever-growing fraction of computer hardware wit be supplied by offshore vendors. Just as solid-state RAMs are used for internal memory, some emerging solid-state technologies could form the basis of memory units in the future. New semiconductor heterostructure devices (superIattices) exhibit electrical, light emitting, and photosensitive properties superior to silicon, which may be applicable to data stor- age. Another solid-state technology with potential data storage ap- plicability is that of silicon-on-insulator (SON) devices for logic. The

OCR for page 14
24 1000 -. . ~ CD 100- ~_ D Ct 10 - - 1 - GL OBAL TRENDS IN COMP UTER TECHNOL OG Y \ , ~ ~ ~ . ..... .. -1 - - .1 - 1985 1990 1995 2000 Year Magnetic Disk Solid State FIGURE 2.2 Storage costs for magnetic disks and solid state. SOURCE: Data are derived and extrapolated from general industry data and trends such as can be found in the Computer Storage Industry Service and the Technical Computer System Industry Service published by Dataquest. higher density chips obtainable with SO] technology could mean more logic available for database operations and potentially larger ~ capacity memories. The submicron complementary metal oxide semiconductor (CMOS) has the potential for increasing the density of [Logic and capacities for RAM, which will mean more logic and larger memories for database operations. The submicron CMOS also offers decreased gate delays and propagation paths, which yield increases in speed. Similarly, GaAs ICs and, in the mid-term, high electron mobility transistor (HEMT) devices will offer high-performance logic, which should yield better access speeds for database operations. Optical Technology. A variety of optical technologies is potentially applicable to data storage- fewtosecond light pulses, ultrahigh-speed fibers, optical logic devices, holograms, and microchannel plates- aTthough Al are currently only at the research stage, and at least 10 years away from development. Nearer term enhancements to conventional optical disk technology (magneto-optics) are surfacing

OCR for page 14
HARD WARE 53 Reading Industry Players The major players in these segments are almost ah U.S. compa- nies. They range from large-scare computer companies (e.g., DEC), to companies focusing on the workstation marketplace (Sun and Apollo), to a range of small start-ups focusing on advanced RISC approaches and personal supercomputers. The United States also leads in the development of multiprocessor technology. While start- ups have played a major role in bringing this technology to market, most established vendors will be supplying multiprocessors in the near future. In both the uniprocessor and multiprocessor arenas, the Euro- peans are consumers of this technology through original-equipment manufacturer and value-added reseller channels; they are not inno- vators. The Japanese have not been major players in these segments, although the recent entry of Sony into the workstation marketplace demonstrates an interest. However, an increasing fraction of the technology used in these machines comes from Japanese IC manufac- turers. If the most advanced IC technology were not made available to U.S. computer companies, the advantage in systems design, archi- tecture, and software that the U.S. companies have could be overcome with higher performance IC technology. Protectability There are three problems in protectability: protecting the basic intellectual content, protecting the ability to replicate, and protecting against the acquisition of small numbers of machines. Although the architectural ideas used in RISC machines and sma]Ll-scaTe vector machines are difficult to protect, experience has shown that a factor of about two in performance can be maintained by the most experienced implementation team. Similarly, while the algorithms for various compiler optimizations will probably be public, the implementation time for an inexperienced group will leave them substantially behind experienced teams. This is an example of the importance of know- how in achieving computer technology advances. In the area of multiprocessing, most of the research work is occurring in university settings, meaning that the results will be public. However, access to these ideas alone does not make it possible to build advanced computers, because the other pieces of the technology may not be accessible.

OCR for page 14
54 GL OBAL TRENDS IN COMP UTER TECHNOL OG Y The ability to build and replicate such machines requires the ability to gain access to advanced IC technology. This could occur either by establishing a manufacturing capability or by buying access to high-speed components and ASICs that would create the ability to build high-speed machines probably capable of running with about 25 percent of the leading-edge technology. In prior times, the performance levels offered by these machines were not so significant that a single machine could make a difference. However, as the machines approach (and even surpass, in the case of a large-scale multicomputer) supercomputer performance, the United States must be more concerned about the potential damage that could be caused by acquisition of even a single machine. The rapid progress in this arena will force the United States to face the issue of whether a single boundary between higher and Tower performance machines can be established to protect access to the higher perfor- mance machines. The alternative is to accept a boundary defined not by absolute performance, but by relation to the fastest machines available. PERSONAL COMPUTERS AND MICROCOMPUTERS While much of the technology assessment in this report addresses the state of the art and expectations for new technology, lower- level technology found in microcomputers plays a key role in the spread of computer technology worldwide and the transition of base technologies into commodity products. This section presents a brief overview of microcomputer technologies to round out the hardware assessment. It covers systems costing under $15,000 (most costing under $7,000) and typically considered personal computers. The more expensive systems are more properly viewed as workstations due to their higher performance. Major Technology Mends PCs are frequently categorized by the type of microprocessor (base technology) they contain, which determines the speed at which they operate and other elements of their performance and function- ality. The key trend in PC performance is the steady progress in microprocessor capability, from the early 8-bit systems introduced in

OCR for page 14
HARD WARE 55 TABLE 2.10 Worldwide Personal Computer Shipments by Type of Microprocessor Intel-Type Micro- Worldwide Shipments (in millions) processors 1981 1982 1983 1984 1985 1986 1987 8088 0.04 0.26 1.08 2.3 2.6 2.7 2.3 8086 0.14 0.35 0.82 1.34 2.1 80286 0.05 0.43 1.3 2.1 80386 0.02 0.21 SOURCE: D ata from G artner Group. the late 1970s to the 32-bit systems now becoming common in busi- ness ant] scientific applications (see Table 2.10~.1 Memory has also been increasing. The price of achieving a given level of performance has been dropping for PCs on the order of 30 percent per year during the mid-l9SOs, and it is expected to continue to fall by about 20 percent per year over the next few years (see Figures 2.6 and 2.7~. Increases in power combined with a growing variety of devices for inputting data for processing and receiving it as output help to support other features that make PCs increasingly easy to use (e.g., graphics displays of ever greater visual resolution and other means for improving the interface with users) and applicable to a growing variety of tasks (see Figure 2.~. The proliferation of PC software discussed in Chapter 4 complements these trends. Progress in performance has accompanied a decade of prolifer- ation of PCs and their uses. While the use of supercomputers has remained essentially confined to the scientific and technical research communities, PCs originally served the home and hobby market, but they have been finding growing uses in scientific and technical, busi- ness and professional, manufacturing, and educational applications. Desktop publishing, database management systems, messaging, and 10Increases in the bit designation signal increases in the size of the chunks of in- formation that can be handled in internal processing and communication steps. They imply more precision in calculations as well as greater speed of operation. Other mea- surements important for assessing the speed and performance capability include clock speed or the speed for performing basic computer operations measured in megahertz (MHz), processing power measured in millions of instructions per second (MIPS), and in numerically intensive applications, the number of floating-point operations per second (Flops).

OCR for page 14
56 14000 1 2000 1 0000 8000 6000 4000 2000 FIGURE 2.6 Declining cost of MIPS for PCs. SOURCE: Courtesy of Gartner Group. 8000 7000 6000 83 5000 PA 4000 3000 GL OBA L TRENDS IN COMP UTER TE CHNOL O G Y 0 . . 1981 1982 1983 1984 1985 1986 1987 1988 1989 9~ 2000 1987 ~ ~~ ~ 1988 al al ~ 1989 Q2 ~ 4 1990 Year FIGURE 2.7 Prices for 386 PCs (with Intel 80386 microprocessors). SOURCE: Courtesy of Gartner Group. ~:~ Other386 PCs IBM 386 Clones IBM 386 PCs other inter- and intra-enterprise communications applications are be- coming more popular, complementing more traditional word process- ing and data processing applications. PCs are expected to become increasingly versatile. In the United States alone, the number of PCs in use grew from approximately 2 million units in 1981 to approximately 45 million units in 198S, and that number is expected to continue to grow at

OCR for page 14
HARD WARE 57 CPU chip set FPU CPU - - MMU 1/0 bus 1 ~ ss bus -l ' ~ 1 1 ~ '1 Communication I Output I Input ~ CRT Act ports devices devices display memory Mass storage (RS-232, (laser (mouse, (output) (RAM) (hard disk) SCSI) _ printer) keyboard) . FIGURE 2.8 One of the many ways of configuring microcomputer components. For example, the bus may be used for both I/O and address; the FPU (floating point unit, an emerging feature) may be included in the CPU; and the communication port may access mass storage. MMU is memory management UDut. SOURCE: Reprinted with permission from Crecine (1986, p. 939~. Copyright 6)1986 by the American Association for the Advancement of Science. a rapid rate over the next five years. Improvements and cost reduc- tions in both hardware and software fuel this PC market growth. Some of the improvements are a product of growing stanciardization. For example, large numbers of applications software packages are written for the few operating systems used by various PCs. Soft- ware is portable or usable among machines sharing an operating system, but the number and differences among operating systems limit that portability. The larger the software base associated with a given hardware-operating system combination, the more valuable the machines involved. A large software base has been a major fac- tor supporting, for example, the popularity of the IBM-PC line and Apple IT and Macintosh lines. Overall, the trend is toward applying PCs to ever more sophis- ticated tasks, including tasks where multiple PCs are interconnected or where PCs are interconnected with larger systems, even super- computers. The trend reflects growth in variety and capability for accessories or peripheral equipment from printers to Moslems, which enable computers to communicate through switched networks (see Chapter 5~. Another trend is an increase in the ability of these sys- tems to perform more than one task at a time. A third trend is the movement of increasing levels of functionality into increasingly

OCR for page 14
58 GL OBA L TRENDS IN COMP UTER TE ClINOL O G Y smaller and smaller machines. The typical PC is a desktop machine, but there is strong growth in smaller, portable, and laptop ma- chines. By 1988, the U.S. laptop PC market was estimated to exceed $1 billion, and it could grow to $2.5 billion in 1989 (Feibus, 1988~. While the RISC architecture discussed in the section on worksta- tions may eventually be applied to PCs, this is not likely to happen until the early l990s. The enormous base of software cleveloped for non-RISC, complex instruction set computing (CISC) architectures and microprocessors and their associated operating systems (e.g., MS/DOS) and the absence of a comparable resource for PC-level DISC machines will slow the transition. Leading Industry Players Early PCs (late 1970s to early 1980s) emanated from U.S. pro- ducers, such as Apple, Tandy/Radio Shack, and Commodore Inter- national, and U.S. producers continue to dominate the market. For example, IBM is estimated to have a 25 to 30 percent share of the PC market overall (Alsop, 1988) and to have supplied 50 percent of PCs used by the U.S. federal government and 23 percent of those used by the U.S. military (Dodge, 1988~. However, the introduction of the IBM-PC (1981), which trig- gered an acceleration in the development and use of PCs, also her- alded an explosion of copy-cat products coming primarily from the Far East, most of them products designed for at least some compat- ibility with the IBM-PC product line (see Figure 2.9~. They have been produced by Japanese, South Korean, and Taiwanese firms in particular; South Korean-made PCs (made by such firms as Hyundai and Samsung) now account for about 10 percent of the U.S. mar- ket. Asian suppliers dominate the Tow end of the market, although the number of suppliers has fallen since about 1986 as a result of Tow-profit margins and a lack of technical differentiation among ma- chines. This lack of differentiation, embodied by the use of common components, allows some Asian manufacturers to achieve economies of scale by producing together PCs sold under a variety of vendor labels (Sanger, 1988~. Asian manufacturers have also been quick to adopt newer, more powerful microprocessors (e.g., Intel 80286 arid 80386~. Interestingly, the internal market in Japan has been slow to develop because of different organization of and attitudes about office work, as well as a shortage of systems capable of handling Japanese language characters.

OCR for page 14
HARD WARE 6- l 5. 4- - o 3 e / I./ -30 a,/: -20 ~ _ 10 _ , , , , . , ~ , . _ 0 1985 1986 1987 1988 1989 1990 1991 1992 1993 FIGURE 2.9 Volumes Ad share of shipments of Asian clone PCs. SOURCE: Courtesy of Gartner Group. 59 Asian Clones es a Pement~t~lPC 5h~ Other countries have also been producing PCs, including not only newly industrializing countries (Brazil, Mexico, and the PRC), but also countries in Western Europe (Italy, the United Kingdom, and France). Production from the Far East, in particular, appears to have been stimulated by the U.S. tendency to produce components or assemble machines in that region. Computers and components through the TBM-PC-AT level (that is, with Intel 80286-type micro- processors) are widely available in this region. Many of the overseas products may lag U.S. designs in terms of quality and level of technol- ogy, but they often have a strong appeal, at least in foreign markets, because of price and local content. This is considered to be the situation in Latin American countries, among others. Foreign markets for PCs are burgeoning (see Figure 2.10~. As early as 1984, the U.S. Department of Commerce was able to esti- mate multimillion-dolIar market sizes and likely growth rates for 30 countries around the world (U.S. Department of Commerce, 1986~. Market research firms estimate that in 1987 more than 2.3 million PCs priced between $600 and $10,000 were shipped to businesses in 17 countries in Western Europe alone (Network Wo rid, August 22, 1988~. About one-quarter of those units were sold by IBM. While the United States may be the largest PC market, its share of world consumption is declining as foreign markets grow.

OCR for page 14
60 GLOBAL TRENDS IN COMPUTER TECHNOLOGY 25 20 15 o ~3 Japan Western Europe E~3 United States 10 _ ~ ::.:' 5 _ //iy, ~L. ~ ,`%j,~ll I. ~ ~ , ;,, me,` `< ~ . _~ ~%, , ,. `, ~ ,. I.. ...... ..... 6 \,\~\ ~ ",'`-~; ,~,,,~, . ,\ ~ .. i ,I,~,~,,,~, -;<,\~7,"~ _,, a'`, `~1~1p mu' 'a',., -~- ~ . 1981 1982 1983 1984 1985 1986 1987 YEAR I, I. 1988 , , %, ,\ . '~::2 I,,, ~ . 1989 1990 , , %, . ~ l 1991 FIGURE 2.10 Estimated annual unit consumption by region (all PCs $0 to $10,000~. SOURCE: Courtesy of Dataquest. Protectability The combination of growing world markets, growing world pro- duction, (including copies or clones as well as indigenous designs), and dependence on base technologies that are commodities produced in a number of countries (not all of them under the CoCom control umbrella) makes the PC the epitome of the commodity computer product. As a result, it is extremely difficult to control the flow of PCs into CMEA countries. The flow has been increasing in the past year as CMEA citizens move to bring PCs back to their countries from tourist and other visits. CMEA countries have been waiving import duties on these machines, and customs officlais in CoCom countries are not always able to differentiate more sophisticated pro- scribed machines from those that can be legally exported (McIntyre, 1988). The growth in and improvement of smaller, portable systems add to the difficulty of protecting this technology. For example, one com- pany has recently announced an innovation in display screens that provides the capacity and clarity typical of a desktop computer in a product clescribed as only slightly bigger than a matchbox (Carroll, 1988~.

OCR for page 14
HARD WARE 61 SYNTHESIS Overall, consumption of computer hardware is rising in CoCom nations. Progress in hardware production and innovation points to some of the interdependency among computer technology elements: machines of different sizes and types may use the same components; development of increasingly sophisticated hardware depends on in- creasingly sophisticated manufacturing and computer-aided design systems (see Chapter 3~; and the application (as wed as develop- ment) of increasingly sophisticated hardware depends on increasingly sophisticated software (see Chapter 4~. Technological evolution may follow a more complex course than is suggested by examining developments in the individual technologies. White many underlying components (or some whole computers) win be commodity items produced in countries with Tow labor costs, it is also likely that the products in which they are embedded wiB have a greater share of their value added at higher levels of their organization. The situation is illustrated by price and performance projections for database management systems (DBMS), which increasingly un- derTie large-scale, high-volume, transaction processing applications, such as banking and reservation systems. As shown in Figure 2.11, performance capacity of high-end transaction processing systems may increase by a factor of more than 100 from 1985 to 2000. While cost per unit capacity may fall by about the same amount, the combination of cost and performance trends would keep the actual dollar cost of high-end systems roughly constant. A growing share of this cost will go not for base components, such as chips, wires, and terminals, but to the very challenging hardware and software engineering problems that must be solved in designing such complex systems. Further, this constant cost will only provide the base hard- ware and DBMS software. An ever-growing additional cost will be devoted to programming and maintenance staff, user-interface soft- ware, and computer-aided software engineering tools to make staff more productive. Competitiveness in producing base components may become less important in the environment described above, but this wiD be true only if there is ready access to whatever quality of components may be needed from overseas producers.

OCR for page 14
62 1 noon - o a: 1 000 cn Q in o ._ - C~ in GL OBAL TRENDS IN COMPUTER TECHNOLOGY 1000 \ 100 10 1 9 8 5 1 9 9 0 1 9 9 5 2 0 0 0 by./ FEW - / - Year cn . 100 ~ En - o ct 10 ~ 0 ? _ 1 3 Transactions per Second K$ per TPS FIGURE 2.11 Speed and cost of database management systems. SOURCE: Data are derived and extrapolated from general industry data and trends such as can be found in the Computer Storage Industry Service and the Tedhnical Computer System Industry Service published by Dataquest. CONCLUSIONS Advances in hardware from semiconductor devices through ma- chines of varying sizes and levels of performance continue to be rapid, with significant improvements in performance taking place in periods from one to three years. This phenomenon accelerates the movement downward of increasing capabilities to computers of various sizes and other computer-controBed devices. NonTeading-edge components and small machines (personal com- puters) have become commodities; they are cheap and are widely available within CoCom and other non-CMEA countries, especially newly industrializing countries in the Far East. The commoditization of hardware makes software increasingly important, both in achieving the full benefit of advances in hardware and in terms of the value of a computer system. Advances in hardware emanate from theory and basic research widely available through the open scientific literature. However,

OCR for page 14
HARD WARE 63 innovation and production strengths depend on the ability to imple- ment new ideas, and that ability requires experience and access to state-of-the-art facilities and infrastructure. High-performance computing is increasingly possible from smaller machines and clusters of smaller machines as wed as larger supercomputers. The trend wiB make controlling high-performance computing increasingly difficult. Supercomputers are a critical technology, and acquisition of only one count have a major impact on a user country. The requirement of these machines for ongoing, labor-intensive support diminishes the value of a possible theft. Nevertheless, it is prudent to protect against improper disposal or transfer of these machines. The risks of diversion in place of supercomputers within CoCom countries may be overstated. This presents a problem for U.S. vendors, who are subject to especially stringent enforcement of regulations covering the installation and operation of these machines. The Uniter! States is a leading innovator in many types of hard- ware, but production leadership is held by or shared with Japan and other countries for commodity hardware and a growing proportion of certain other types of hardware.