Advances in distributed sensor technology

regarding computational science in undergrad­ inar). For the same reasons, I have difficulty uate and graduate education. The book closes imagining it being used for self-instruction by with references to the technical reports prepared someone who does not know much about par­ by Caltech research team members, their allel computing. biographies, and an extended bibliography. This book can be very valuable to special­ From this short description of the book's ists, but even for this audience it has draw­ contents, it should be clear that the authors backs. Because the book reports only on Cal­ have gathered a substantial body of evidence tech research, it gives a rather one-sided to demonstrate that parallel computing really picture of most issues. This is not a problem works and can be extremely useful in solving when it discusses examples of successfully large, computationally intensive problems in implemented problems and tools. This does a number of disciplines. From this point of become a problem, though, when on the basis view, this book is an important contribution of the research done in one place, the authors to the current discussions of the role of high- make more general claims about the state and performance computing. This role is under the future of parallel computing. reevaluation because of the shrinking US fed­ Finally, even though only three names eral science budget and the spectacular bank­ appear on the book's cover, almost 30 co­ ruptcies and buyouts that mark the last two authors—members of the Caltech research years in the industry. group—wrote various sections of the book. At the same time, the book left me with This gives rise to a rather uneven writing mixed feelings. My initial question is, who was style. While most of the book is written very the target audience? Because the book is clearly, some sections are written in a rather extremely broad in scope, it requires a rather awkward style, and some sections unneces­ extensive background in mathematics, physics sarily repeat material from other places. The (including physical chemistry), and computer strangest section is 6.6, which describes char­ science. This requirement is understandable acter recognition using neural networks—it and reasonable because of the subject's inter­ does not mention parallel computing even disciplinary nature. However, because of the once. breadth of the presentation, there is no place to Parallel Computing Works! is a rather specific provide additional basic information before book. However, I recommend it to computa­ presenting the more advanced subjects (thus tional science specialists. It shows, through making the book more accessible). So, this examples, that parallel computing really works, book clearly cannot be used as a textbook and it presents the tools and techniques that (except possibly for an advanced graduate sem­ make it happen. Advances in Distributed Sensor Technology reviewed by Jerry D. Cavin, University of Texas at Austin Advances in Distributed Sensor Technology is many disciplines. The book gives many real- Advances in Distributed based on earlier research papers the authors life examples of the types of problems that Sensor Technology wrote for the US Office of Naval Research. involve sensors, such as edge detection, object by S.S. lyengar, L. Prasad, and The authors claim to have revised and recognition, and aerospace and defense sensing. Hla Min expanded the material to target the book for The authors supply numerous definitions 273 pp. senior-level undergraduate or introductory during the discussion of sensor integration. $68 graduate courses. However, their definition of "sensor integra­ Prentice Hall The authors assume that the reader is new tion" is confusing. They define it as "the process Upper Saddle River, N.J. to distributed sensor technology, and a great of integrating, combining, or fusing the sensor 1995 deal of the book builds the foundation of readings." They continue to use both "sensor ISBN 0-13-360033-5 mathematics and definitions required to integration" and "sensor fusion" throughout the understand the later sections. The book starts text and figures. At some points in the text, I with a tutorial covering the essentials of areas wondered whether they were talking about two of mathematics important to distributed sen­ distinctly different things. A more consistent use sors, such as set theory, relations, functions, of a single term would be less confusing. and graph theory. There is an excellent dis­ The sections of the book covering sensor cussion of the range of sensors currently in fusion for single-dimensional and multi­ use, and of how the constantly evolving dimensional sensors are well written. They designs of sensor systems have branched into step the reader through complex algorithms Winter 1996 83  with pseudocode, giving the mathematics tion in distributed systems. This topic is behind the algorithms. This part is clearly the cetainly worth more discussion than the main focus, around which the entire book is authors provide. built. The sections on distributed routing The subject of distributed sensor technol­ techniques are also very well explained. The ogy covers a wide area. Covering all topics authors give algorithms to connect the sen­ within the confines of a single book is a for­ sors into distributed networks. The discussion midable task—one at which this book fails. For on temporal and fault-tolerance issues of the a classroom text, many parts of the book treat different network architectures is excellent. their subject far too generally. The introduc­ This book's title led me to believe the book tory material provides only a basic foundation. would discuss many of the key areas of dis­ Many topics will need to be supplemented for tributed sensor technology where new and classroom use if students are to grasp the big important breakthroughs have occurred. But picture of distributed sensor technology. the book barely mentions many areas impor­ Advances in Distributed Sensor Technology tant to distributed sensors, and does not dis­ would not make a good textbook, but as an expe­ cuss others at all. For example, it mentions rienced engineer, I found it an excellent source clock synchronization only briefly and fails to of information on sensor fusion, distributed include any algorithms for time synchroniza­ routing, and sensor-modeling algorithms. Fault-Tolerant Computer System Design reviewed by Niraj K.Jha, Princeton University The fault-tolerant computingfieldis about four Chapter 2 describes ten fault-tolerant archi­ Fault-Tolerant Computer decades old. It recently celebrated the silver tectures, logically arranged under three cate­ System Design anniversary of the premier Symposium on gories, depending on the intended application: by Dhiraj K. Pradhan Fault-Tolerant Computing. Nearly every issue high-availability systems, long-life systems, 550 pp. of IEEE Transactions on Computers and Transac­and critical computations. Chapter 3 covers $72 tions on Parallel and Distributed Systems, among fault-tolerant multiprocessor and distributed Prentice Hall many other journals, has at least one paper on architectures. The discussion of fault toler­ Upper Saddle River, NJ. fault tolerance. Thus, this book is very timely. ance in distributed architectures—especially 1996 This book's predecessor, Fault-Tolerant the latest forward-recovery schemes—is very ISBN 0-13-057887-8 Computing: Theory and Techniques, published a timely. This chapter also discusses fault toler­ decade ago, was very well received. It dealt ance in various interconnection networks. with the architecture of fault-tolerant com­ Chapter 4 complements Chapter 3 nicely; it puters, fault-tolerant multiprocessors, system provides case studies of fault-tolerant multi­ diagnosis, reliability estimation, and fault- processor and distributed systems. tolerant software. The new book is both Chapter 5 shows how you can experimen­ broader and deeper. The additional topics are tally analyze the dependability of computer fault tolerance in distributed systems and systems. It describes various appropriate sta­ experimental analysis of computer system tistical techniques and provides techniques dependability. Also, the book covers each of that apply to each of these phases: design, pro­ the previous topics in much greater detail. totype, and operational. Chapter 6 deals with This book is suitable as a textbook for gradu­ traditional refiabifity-estimation techniques. ate-level electrical engineering or computer sci­ Chapter 8 covers system-level diagnosis tech­ ence courses. The ends of chapters feature some niques under both the bounded and proba­ interesting problems ofmedium difficulty, which bilistic models. an instructor mightfindvery useful. However, An important additional chapter, Chapter because of the nature of the topics, not every 7, deals with fault-tolerant software. This chapter has a large problem set This book will includes discussion of design diversity, soft­ also be very useful to practicing engineers. ware reliability models, acceptance tests, and The book is logically organized. The first exception handling. chapter introduces the fundamentals of fault Fault-Tolerant Computer System Design is tolerance, such as different types of redundancy reasonably comprehensive and self-contained. techniques and dependability-evaluation tech­ There is sufficient continuity within and niques. Chapters 2, 3, 4, and 7 discuss the among chapters. Thefiguresare not as appeal­ design of fault-tolerant systems, and Chapters ing as they could have been; however, this does 5,6, and 8 cover the analysis of such systems. not detract muchfromthe book's logical flow. 84 IEEE Parallel & Distributed Technology