Browse Books

Reviewer: David Bruce Henderson

Much of modern society depends on fast, inexpensive computers and on the performance of these computers continuing to rapidly increase, just as they have for the past 50 years. This report from the Committee on Sustaining Growth in Computing Performance of the National Research Council of the National Academy of Science documents the plateauing of growth in computing performance around 2004 and discusses what this means for the future of computing. Chapter 1 discusses the widespread expectation that computer performance will continue to improve and that computer costs will continue to fall. The discussion covers why faster computers are needed and the importance of this performance growth for the sciences, defense, business, and the average citizen. Chapter 2 explains just what is meant by computer "performance," and chapter 3 looks in some detail at the fundamental reasons why performance has now plateaued. Several factors at the crux of this sudden halt in performance growth, such as heat density and fabrication complexity, are discussed. Chapter 4 examines the possibility of restarting growth in performance through parallel processing in both software and hardware. Parallelism is not new, but as the report notes, no parallel solution has yet delivered on the promise of significant performance improvement. Perhaps it has just been easier to rely on the continued growth of chip performance. Chapter 5 concludes with the committee's recommendations on the direction of future research to potentially restart growth in computer performance. The report contains a preface, a table of contents, and an abstract-but no index. The detail in the table of contents, however, is adequate for locating the various topics throughout the report. Clear diagrams, footnotes, and boxed discussions of relevant side topics punctuate the report. A summary chapter is included, which highlights the major findings and recommendations. For the reader with little spare time, simply reading the three pages of the abstract and the 16-page summary is all that is really needed to grasp the committee's findings. The report has several appendices covering historical aspects of computer performance, biographies of those involved in the production of the report, and reprints of relevant earlier papers. Notable is a reprint of Gordon Moore's 1965 paper [1], which led to the phrase "Moore's law," commonly used to describe the exponential growth of computer performance since the 1950s. The report is quite readable and should provide useful guidance on the future direction of research into improvements in computer performance. The summary at the beginning of the report provides a good executive brief of the findings and recommendations. Whether computer performance growth will ever return to the dramatic rates of the past 50 years remains to be seen. Online Computing Reviews Service