For a mature, established discipline, a handbook is often a comprehensive repository of proven knowledge that a practitioner uses as a reference, or perhaps as a source of ideas for brainstorming. Parallel computing is, however, anything but a mature, established discipline. What can the ambitious moniker of “handbook” propose in this case? This review addresses the question from the perspective of a diligent practitioner with a penchant for academic erudition.

This ponderous opus weighs in at more than 1,200 pages (and almost five pounds). The book’s stated audience includes “researchers, developers, educators, and students.” It is a compilation of 47 monographs, authored by more than 100 contributors, and grouped in three broad categories: “Models,” “Algorithms,” and “Applications.” “Models” deals broadly with theoretical archetypes of parallel computation. “Algorithms” includes surveys and original works on distributed algorithms. “Applications” is a varied collection of essays on problems and solutions related to parallel systems.

The content varies from the exotic to the almost immediately applicable, with only a few examples of the extremes. Some monographs tax the reader to the extent that only the truly interested will be prepared to spend the effort necessary for understanding, whereas a few are almost ready for immediate consumption by a practitioner who may be tackling a relevant problem. The bulk of the content is in the theoretically understandable middle: information that is valuable, provided one can find a manner to put it to practice.

What is missing, though, is a thread to provide continuity in the presentation of subjects. The preface spends half a dozen pages introducing a presumptive path of evolution for a collection of monographs that seems otherwise assembled without a precise focus: Breadth? Depth? What else? The result is a “handbook” with a prominent identity crisis: it is neither a tutorial, nor a reference, nor a snapshot of the state of the art.

On one extreme, there is no comprehensive, structured explanation of why parallel computing may be a good idea, and sometimes even necessary. Seldom is it even cursorily mentioned that parallel computing is appropriate when the problem is too complex, or the data set too large, for sequential processing. For the readership of this book, the need for parallel computing is taken for granted, as is all of the basic information and most of the not so basic.

On the other extreme, for instance, there is no mention of transactional memory, a just-about-new idea, now under development for fine-granularity synchronization in shared-memory environments. Transactional memory was not unknown at the time of this book’s publication.

On the bright side, the bibliography of most monographs is detailed, rich, and more often than not fairly recent. These references can be inspiring to anyone looking for greater depth on the subject. Of course, there is also the occasional paper with no references newer than the 20th century. This is not surprising, given the unevenness of the handbook’s tone. The writing style is uneven as well. To a degree this is to be expected, given the number of contributors. It is, however, infuriatingly uneven, to the point of suspecting substandard editing, especially considering the presence of several spelling errors.

Alas, few will read this opus from cover to cover. Some will find a few of the monographs to be subjects of interest, and more will find the good literature surveys worthy of attention. The contribution of this work will be in inspiring a researcher, in helping prepare a few classroom lectures, and in supporting the occasional practitioner dealing with a focused problem like, for instance, designing a competent distributed scheduler for, say, a simple object access protocol (SOAP) load balancer. The places where this tome will most likely be found are on the shelves of a few professorial or near-professorial bookcases, or, more likely, in the stacks of a library for occasional consultation.