If... history... teaches us anything, it is that man in his quest for knowledge and progress, is determined and cannot be deterred.
John F. Kennedy, Address at Rice University (1962) A section of historical perspectives closes each chapter in the text. This section provides historical background on some of the key ideas presented in the chapter.
The authors may trace the development of an idea through a series of machines or describe significant projects. If you’re interested in examining the initial develop-ment of an idea or machine or interested in further reading, references are provided at the end of the section. In this historical section, we discuss the early development of digital computers and the development of performance measurement methodol-ogies. The development of the key innovations in desktop, server, and embedded processor architectures are discussed in historical sections in virtually every chapter of the book.
1.11 Historical Perspective and References
The First General-Purpose Electronic Computers
J. Presper Eckert and John Mauchly at the Moore School of the University of Pennsylvania built the world’s first fully-operational electronic general-purpose computer. This machine, called ENIAC (Electronic Numerical Integrator and Calculator), was funded by the U.S. Army and became operational during World War II, but it was not publicly disclosed until 1946. ENIAC was used for comput-ing artillery fircomput-ing tables. The machine was enormous—100 feet long, 8 1/2 feet high, and several feet wide. Each of the 20 10-digit registers was 2 feet long. In total, there were 18,000 vacuum tubes.
Although the size was three orders of magnitude bigger than the size of the av-erage machines built today, it was more than five orders of magnitude slower, with an add taking 200 microseconds. The ENIAC provided conditional jumps and was programmable, which clearly distinguished it from earlier calculators.
Programming was done manually by plugging up cables and setting switches and required from a half-hour to a whole day. Data were provided on punched cards.
The ENIAC was limited primarily by a small amount of storage and tedious pro-gramming.
In 1944, John von Neumann was attracted to the ENIAC project. The group wanted to improve the way programs were entered and discussed storing programs as numbers; von Neumann helped crystallize the ideas and wrote a memo proposing a stored-program computer called EDVAC (Electronic Discrete Variable Automatic Computer). Herman Goldstine distributed the memo and put von Neumann’s name on it, much to the dismay of Eckert and Mauchly, whose names were omitted. This memo has served as the basis for the commonly used term von Neumann computer. Several early inventors in the computer field be-lieve that this term gives too much credit to von Neumann, who conceptualized and wrote up the ideas, and too little to the engineers, Eckert and Mauchly, who worked on the machines. Like most historians, your authors (winners of the 2000 IEEE von Neumann Medal) believe that all three individuals played a key role in developing the stored program computer. von Neumann’s role in writing up the ideas, in generalizing them, and in thinking about the programming aspects was critical in transferring the ideas to a wider audience.
In 1946, Maurice Wilkes of Cambridge University visited the Moore School to attend the latter part of a series of lectures on developments in electronic com-puters. When he returned to Cambridge, Wilkes decided to embark on a project to build a stored-program computer named EDSAC, for Electronic Delay Storage Automatic Calculator. (The EDSAC used mercury delay lines for its memory;
hence the phrase “delay storage” in its name.) The EDSAC became operational in 1949 and was the world’s first full-scale, operational, stored-program computer [Wilkes, Wheeler, and Gill 1951; Wilkes 1985, 1995]. (A small prototype called the Mark I, which was built at the University of Manchester and ran in 1948, might be called the first operational stored-program machine.) The EDSAC was an accumulator-based architecture. This style of instruction set architecture
re-mained popular until the early 1970s. (Chapter 2 starts with a brief summary of the EDSAC instruction set.)
In 1947, Eckert and Mauchly applied for a patent on electronic computers.
The dean of the Moore School, by demanding the patent be turned over to the university, may have helped Eckert and Mauchly conclude they should leave.
Their departure crippled the EDVAC project, which did not become operational until 1952.
Goldstine left to join von Neumann at the Institute for Advanced Study at Princeton in 1946. Together with Arthur Burks, they issued a report based on the 1944 memo [1946]. The paper led to the IAS machine built by Julian Bigelow at Princeton’s Institute for Advanced Study. It had a total of 1024 40-bit words and was roughly 10 times faster than ENIAC. The group thought about uses for the machine, published a set of reports, and encouraged visitors. These reports and visitors inspired the development of a number of new computers, including the first IBM computer, the 701, which was based on the IAS machine. The paper by Burks, Goldstine, and von Neumann was incredible for the period. Reading it to-day, you would never guess this landmark paper was written more than 50 years ago, as most of the architectural concepts seen in modern computers are dis-cussed there (e.g., see the quote at the beginning of Chapter 5).
In the same time period as ENIAC, Howard Aiken was designing an electro-mechanical computer called the Mark-I at Harvard. The Mark-I was built by a team of engineers from IBM. He followed the Mark-I by a relay machine, the Mark-II, and a pair of vacuum tube machines, the Mark-III and Mark-IV. The Mark-III and Mark-IV were being built after the first stored-program machines.
Because they had separate memories for instructions and data, the machines were regarded as reactionary by the advocates of stored-program computers. The term Harvard architecture was coined to describe this type of machine. Though clear-ly different from the original sense, this term is used today to appclear-ly to machines with a single main memory but with separate instruction and data caches.
The Whirlwind project [Redmond and Smith 1980] began at MIT in 1947 and was aimed at applications in real-time radar signal processing. Although it led to several inventions, its overwhelming innovation was the creation of magnetic core memory, the first reliable and inexpensive memory technology. Whirlwind had 2048 16-bit words of magnetic core. Magnetic cores served as the main memory technology for nearly 30 years.
Important Special-Purpose Machines
During the Second Wold War, there were major computing efforts in both Great Britain and the United States focused on special-purpose code-breaking comput-ers. The work in Great Britain was aimed at decrypting messages encoded with the German Enigma coding machine. This work, which occurred at a location called Bletchley Park, led to two important machines. The first, an electrome-chanical machine, conceived of by Alan Turing, was called BOMB [see Good in
Metropolis 1980]. The second, much larger and electronic machine, conceived and designed by Newman and Flowers, was called COLOSSUS [see Randall in Metropolis 1980]. These were highly specialized cryptanalysis machines, which played a vital role in the war by providing the ability to read coded messages, es-pecially those sent to U-boats. The work at Bletchley Park was highly classified (indeed some of it is still classified), and, so, its direct impact on the development of ENIAC, EDSAC and other computers is hard to trace, but it certainly had an indirect effect in advancing the technology and gaining understanding of the is-sues.
Similar work on special-purpose computers for cryptanalysis went on in the United States. The most direct descendent of this effort was a company Engineer-ing Research Associates (ERA [see Thomash in Metropolis 1980], which was founded after the war to attempt to commercialize on the key ideas. ERA build several machines, which were sold to secret government agencies, and was even-tually purchased by Sperry Rand, which had earlier purchased the Eckert Mauch-ly Computer Corporation.
Another early set of machines that deserves credit was a group of special-pur-pose machines built by Konrad Zuse in Germany in the late 1930s and early 1940s [see Bauer and Zuse in Metropolis 1980]. In addition to producing an op-erating machine, Zuse was the first to implement floating point, which von Neu-mann claimed was unnecessary!. His early machines used a mechanical store that was smaller than other electromechanical solutions of the time. His last machine was electromechanical but, because of the war, never completed.
An important early contributor to the development of electronic computers was John Atanasoff, who built a small-scale electronic computer in the early 1940s [Atanasoff 1940]. His machine, designed at Iowa State University, was a special-purpose computer (called the ABC: Atanasoff Berry Computer) that was never completely operational. Mauchly briefly visited Atanasoff before he built ENIAC and several of Atansanoff’s ideas (e.g. using binary representation) likely influenced Mauchly. The presence of the Atanasoff machine, together with delays in filing the ENIAC patents (the work was classified and patents could not be filed until after the war) and the distribution of von Neumann’s EDVAC paper, were used to break the Eckert-Mauchly patent [Larson 1973]. Though controver-sy still rages over Atanasoff’s role, Eckert and Mauchly are usually given credit for building the first working, general-purpose, electronic computer [Stern 1980].
Atanasoff, however, demonstrated several important innovations included in later computers. Atanasoff deserves much credit for his work, and he might fairly be given credit for the world’s first special-purpose electronic computer and for pos-sibly influencing Eckert and Mauchly.
Commercial Developments
In December 1947, Eckert and Mauchly formed Eckert-Mauchly Computer Cor-poration. Their first machine, the BINAC, was built for Northrop and was shown
in August 1949. After some financial difficulties, the Eckert-Mauchly Computer Corporation was acquired by Remington-Rand, later called Rand. Sperry-Rand merged the Eckert-Mauchly acquisition, ERA, and its tabulating business to form a dedicated computer division, called UNIVAC. UNIVAC delivered its first computer, the UNIVAC I in June 1951. The UNIVAC I sold for $250,000 and was the first successful commercial computer—48 systems were built! To-day, this early machine, along with many other fascinating pieces of computer lore, can be seen at the Computer Museum in Mountain View, California. Other places where early computing systems can be visited include the Deutsches Mu-seum in Munich, and the Smithsonian in Washington, D.C., as well as numerous online virtual museums.
IBM, which earlier had been in the punched card and office automation busi-ness, didn’t start building computers until 1950. The first IBM computer, the IBM 701 based on von Neumann’s IAS machine, shipped in 1952 and eventually sold 19 units [see Hurd in Metropolis 1980].In the early 1950s, many people were pessimistic about the future of computers, believing that the market and op-portunities for these “highly specialized” machines were quite limited. Nonethe-less, IBM quickly became the most successful computer company. The focus on reliability and a customer and market driven strategy was key. Although the 701 and 702 were modest successes, IBM’s next machine the 704/705, first delivered in 1954, greatly exceeded its initial sales forecast of 50 machines, thanks in part to the inclusion of core memory.
Several books describing the early days of computing have been written by the pioneers [Wilkes 1985, 1995; Goldstine 1972], as well as [Metropolis, Howlett, and Rota 1980], which is a collection of recollections by early pioneers. There are numerous independent histories, often built around the people involved [Slat-er 1987], as well as a journal, Annals of the History of Computing, devoted to the history of computing.
The history of some of the computers invented after 1960 can be found in Chapter 2 (the IBM 360, the DEC VAX, the Intel 80x86, and the early RISC machines), Chapters 3 and 4 (the pipelined processors, including Stretch and the CDC 6600), and Appendix B (vector processors including the TI ASC, CDC Star, and Cray processors).
Development of Quantitative Performance Measures:
Successes and Failures
In the earliest days of computing, designers set performance goals—ENIAC was to be 1000 times faster than the Harvard Mark-I, and the IBM Stretch (7030) was to be 100 times faster than the fastest machine in existence. What wasn’t clear, though, was how this performance was to be measured. In looking back over the years, it is a consistent theme that each generation of computers obsoletes the performance evaluation techniques of the prior generation.
The original measure of performance was time to perform an individual oper-ation, such as addition. Since most instructions took the same execution time, the timing of one gave insight into the others. As the execution times of instructions in a machine became more diverse, however, the time for one operation was no longer useful for comparisons. To take these differences into account, an instruc-tion mix was calculated by measuring the relative frequency of instrucinstruc-tions in a computer across many programs. The Gibson mix [Gibson 1970] was an early popular instruction mix. Multiplying the time for each instruction times its weight in the mix gave the user the average instruction execution time. (If mea-sured in clock cycles, average instruction execution time is the same as average CPI.) Since instruction sets were similar, this was a more accurate comparison than add times. From average instruction execution time, then, it was only a small step to MIPS (as we have seen, the one is the inverse of the other). MIPS had the virtue of being easy for the layman to understand.
As CPUs became more sophisticated and relied on memory hierarchies and pipelining, there was no longer a single execution time per instruction; MIPS could not be calculated from the mix and the manual. The next step was bench-marking using kernels and synthetic programs. Curnow and Wichmann [1976]
created the Whetstone synthetic program by measuring scientific programs writ-ten in Algol 60. This program was converted to FORTRAN and was widely used to characterize scientific program performance. An effort with similar goals to Whetstone, the Livermore FORTRAN Kernels, was made by McMahon [1986]
and researchers at Lawrence Livermore Laboratory in an attempt to establish a benchmark for supercomputers. These kernels, however, consisted of loops from real programs.
As it became clear that using MIPS to compare architectures with different in-structions sets would not work, a notion of relative MIPS was created. When the VAX-11/780 was ready for announcement in 1977, DEC ran small benchmarks that were also run on an IBM 370/158. IBM marketing referred to the 370/158 as a 1-MIPS computer, and since the programs ran at the same speed, DEC market-ing called the VAX-11/780 a 1-MIPS computer. Relative MIPS for a machine M was defined based on some reference machine as
The popularity of the VAX-11/780 made it a popular reference machine for rela-tive MIPS, especially since relarela-tive MIPS for a 1-MIPS computer is easy to calculate: If a machine was five times faster than the VAX-11/780, for that bench-mark its rating would be 5 relative MIPS. The 1-MIPS rating was unquestioned for four years, until Joel Emer of DEC measured the VAX-11/780 under a time-sharing load. He found that the VAX-11/780 native MIPS rating was 0.5. Subse-quent VAXes that run 3 native MIPS for some benchmarks were therefore called
MIPSM PerformanceM Performancereference
---×MIPSreference
=
6-MIPS machines because they run six times faster than the VAX-11/780. By the early 1980s, the term MIPS was almost universally used to mean relative MIPS.
The 1970s and 1980s marked the growth of the supercomputer industry, which was defined by high performance on floating-point-intensive programs. Average instruction time and MIPS were clearly inappropriate metrics for this industry, hence the invention of MFLOPS (Millions of FLoating-point Operations Per Sec-ond), which effectively measured the inverse of execution time for a benchmark. . Unfortunately customers quickly forget the program used for the rating, and mar-keting groups decided to start quoting peak MFLOPS in the supercomputer per-formance wars.
SPEC (System Performance and Evaluation Cooperative) was founded in the late 1980s to try to improve the state of benchmarking and make a more valid ba-sis for comparison. The group initially focused on workstations and servers in the UNIX marketplace, and that remains the primary focus of these benchmarks to-day. The first release of SPEC benchmarks, now called SPEC89, was a substan-tial improvement in the use of more realistic benchmarks.
References
AMDAHL, G. M. [1967]. “Validity of the single processor approach to achieving large scale comput-ing capabilities,” Proc. AFIPS 1967 Sprcomput-ing Joint Computer Conf. 30 (April), Atlantic City, N.J., 483–485.
ATANASOFF, J. V. [1940]. “Computing machine for the solution of large systems of linear equations,”
Internal Report, Iowa State University, Ames.
BELL, C. G. [1984]. “The mini and micro industries,” IEEE Computer 17:10 (October), 14–30.
BELL, C. G., J. C. MUDGE,AND J. E. MCNAMARA [1978]. A DEC View of Computer Engineering, Digital Press, Bedford, Mass.
BURKS, A. W., H. H. GOLDSTINE,AND J. VON NEUMANN [1946]. “Preliminary discussion of the logi-cal design of an electronic computing instrument,” Report to the U.S. Army Ordnance Department, p. 1; also appears in Papers of John von Neumann, W. Aspray and A. Burks, eds., MIT Press, Cam-bridge, Mass., and Tomash Publishers, Los Angeles, Calif., 1987, 97–146.
CURNOW, H. J. AND B. A. WICHMANN [1976]. “A synthetic benchmark,” The Computer J., 19:1.
FLEMMING, P. J. AND J. J. WALLACE [1986]. “How not to lie with statistics: The correct way to summarize benchmarks results,” Comm. ACM 29:3 (March), 218–221.
FULLER, S. H. AND W. E. BURR [1977]. “Measurement and evaluation of alternative computer architectures,” Computer 10:10 (October), 24–35.
GIBSON, J. C. [1970]. “The Gibson mix,” Rep. TR. 00.2043, IBM Systems Development Division, Poughkeepsie, N.Y. (Research done in 1959.)
GOLDSTINE, H. H. [1972]. The Computer: From Pascal to von Neumann, Princeton University Press, Princeton, N.J.
JAIN, R. [1991]. The Art of Computer Systems Performance Analysis: Techniques for Experimental Design, Measurement, Simulation, and Modeling, Wiley, New York.
LARSON, E. R. [1973]. “Findings of fact, conclusions of law, and order for judgment,” File No. 4–67, Civ. 138, Honeywell v. Sperry Rand and Illinois Scientific Development, U.S. District Court for the State of Minnesota, Fourth Division (October 19).
LUBECK, O., J. MOORE,AND R. MENDEZ [1985]. “A benchmark comparison of three supercomputers:
Fujitsu VP-200, Hitachi S810/20, and Cray X-MP/2,” Computer 18:12 (December), 10–24.
METROPOLIS, N., J. HOWLETT,AND G-C ROTA,EDITORS [1980], A History of Computing in the Twentieth Century, Academic Press, N.Y.
MCMAHON, F. M. [1986]. “The Livermore FORTRAN kernels: A computer test of numerical perfor-mance range,” Tech. Rep. UCRL-55745, Lawrence Livermore National Laboratory, Univ. of Cali-fornia, Livermore (December).
REDMOND, K. C. AND T. M. SMITH [1980]. Project Whirlwind—The History of a Pioneer Computer, Digital Press, Boston.
SHURKIN, J. [1984]. Engines of the Mind: A History of the Computer, W. W. Norton, New York.
SLATER, R. [1987]. Portraits in Silicon, MIT Press, Cambridge, Mass.
SMITH, J. E. [1988]. “Characterizing computer performance with a single number,” Comm. ACM 31:10 (October), 1202–1206.
SPEC [1989]. SPEC Benchmark Suite Release 1.0, October 2, 1989.
SPEC [1994]. SPEC Newsletter (June).
STERN, N. [1980]. “Who invented the first electronic digital computer,” Annals of the History of Computing 2:4 (October), 375–376.
TOUMA, W. R. [1993]. The Dynamics of the Computer Industry: Modeling the Supply of Work-stations and Their Components, Kluwer Academic, Boston.
WEICKER, R. P. [1984]. “Dhrystone: A synthetic systems programming benchmark,” Comm. ACM 27:10 (October), 1013–1030.
WILKES, M. V. [1985]. Memoirs of a Computer Pioneer, MIT Press, Cambridge, Mass.
WILKES, M. V. [1995]. Computing Perspectives, Morgan Kaufmann, San Francisco.
WILKES, M. V., D. J. WHEELER,AND S. GILL [1951]. The Preparation of Programs for an Electronic Digital Computer, Addison-Wesley, Cambridge, Mass.
E X E R C I S E S
Each exercise has a difficulty rating in square brackets and a list of the chapter sections it depends on in angle brackets. See the Preface for a description of the difficulty scale.
still a good exercise
1.1 [20/10/10/15] <1.6> In this exercise, assume that we are considering enhancing a ma-chine by adding a vector mode to it. When a computation is run in vector mode it is 20 times
1.1 [20/10/10/15] <1.6> In this exercise, assume that we are considering enhancing a ma-chine by adding a vector mode to it. When a computation is run in vector mode it is 20 times