Browse Books

Reviewer: David Malcolm Nicol

Solving problems on concurrent processors is a long-awaited outgrowth of the extensive experience at the California Institute of Technology in the field of solving scientific problems on message-passing parallel processors. The book shows how many standard mathematical operations can be distributed, including iterative relaxation methods, finite-element methods, common matrix operations, FFTs, simulated annealing, particle dynamics, and circuit simulations. Simple analyses that describe the overhead cost of parallelism accompany the examples. These analyses serve to demonstrate that the problem decompositions can be expected to scale well onto much larger parallel computers. The book is sprinkled with side discussions about CUBIX, the operating system and user interface developed on the Cal Tech hypercube. The authors are physicists, and operating system purists may sniff at CUBIX. The system interface presented to a programmer is quite practical, though, even if unexciting. A person with a baccalaureate degree in a hard science will have sufficient background to read this book. Anyone interested in practical parallel computing must read it. I intend to use the book in a first-year graduate course on parallel computing. Many of the programs described in the book are sketched in pseudocode; most of these programs (and the operating system) have been placed in the public domain and are available for a nominal fee.