Hord’s survey of multiple instruction multiple data (MIMD) parallel processing technology begins with a brief background section intended to supply the reader with some basic MIMD concepts and to describe the motivation behind the design of this class of computers. This introduction is followed by 23 independent chapters organized into three sections. The first section, “MIMD Computers,” comprises 15 chapters, each devoted to a single MIMD machine. Of the machines surveyed, 11 are commercial systems while the remaining 4 were developed for research purposes. The survey style of presentation is retained in the last 8 chapters, which are organized into two major sections entitled “MIMD Software” and “MIMD Issues.” Together, these two sections present specific examples of programming environments and techniques related to MIMD processing.

The commercial systems surveyed in the opening chapters generally constitute a representative cross-section of contemporary MIMD machines, with the notable exception of any member of the Cray family of multiprocessors. Similarly, the inclusion of New York University’s Ultracomputer and the University of Illinois’s Cedar system might have made the author’s coverage of research machines more comprehensive. In addition to such monolithic parallel architectures, Hord appropriately introduces heterogeneous networks of supercomputers as a viable form of MIMD processing, albeit not until the final chapter.

Readers new to parallel processing may be misled by the awkward organization of Section 1. The survey begins with a discussion of Thinking Machines’s CM-5, a machine that has often been classified as a single program multiple data rather than MIMD architecture. While one may debate this distinction, the CM-5’s hybrid design does not serve well as a first example for the uninitiated. The author further confuses the issue with statements such as “the CM-5 architecture is designed to support especially well the data parallel model of programming” and by making reference to “the CM-5 data parallel architecture.” Older machines that might have made better initial examples, such as the Encore Multimax and Sequent Symmetry, do not appear until the sixth and tenth chapters, respectively. Also, Hord makes no attempt to group chapters describing machines with similar organizations, such as shared memory and message-passing machines.

The next two sections address some important issues and developments in parallel programming and systems software. Specific examples of a parallel operating system  (Trollius),  programming languages (Apply and Linda), and tools (PTOOL, PVW, and Hence) are presented in chapters 16, 17, 18, 19, and 23. While these chapters provide a sampling of contemporary parallel software, others seem to digress from the subject, such as chapter 21, “Mathematical Model Partitioning and Packing for Parallel Computer Calculation.” It is inevitable that the selection of material for these two sections should seem eclectic, given the limited space allocated to subjects as broad as “MIMD Software” and “MIMD Issues.”

In general, this book lacks the organization and cohesion necessary for a good survey. Most notably, the chapters of Section 1 are inconsistent in their treatment of the various MIMD machines. Bibliographies, programming examples, discussions of I/O issues, and performance figures are present (to varying degrees) in some chapters while totally absent in others, leaving the reader with little upon which to base any meaningful comparisons. Similar observations can be made about Sections 2 and 3.

This lack of consistency can be largely attributed to the diverse origins of the book’s contents for, as the author points out in the foreword, “Much of the material included here is based on or adapted from work previously published in government reports, journal articles, books and vendor literature.” Several entire chapters and portions of other chapters are taken from these sources. This cut-and-paste style of editing is distracting and tends to produce redundant information, sometimes within the space of a few pages. To add to the reader’s confusion, the origins of these contributions often are not referenced in the chapters in which the contributions appear and are instead given in a “List of Sources” at the book’s end.

In a work that draws heavily from other sources, much depends on the selection of the borrowed material. The quality of these contributions varies widely, however, from well-established research papers to low-content promotional literature. Generally, chapters covering newer MIMD multiprocessors fare worse in this regard, presumably due to a shortage of published research conducted on these machines.

Although intended for a “wide range of readers,” this book is suitable only for professionals already familiar with MIMD processing. Students and those looking to expand their understanding of parallel processing technology will quickly become confused by the book’s lack of organization and often obfuscated presentations. While Hord’s book is perhaps the most current and exhaustive survey of MIMD parallel computers, readers searching for a meaningful and coherent treatment of this topic should look elsewhere.