Automatic Algorithm Recognition and Replacement: A New Approach to Program Optimisation

Automatic Algorithm Recognition and Replacement: A New Approach to Program Optimisation

Robert Metzger and Zhaofang WenMIT PressCambridge, MA219 pp.ISBN 0-262-13368-7$40

Keywords: Publication, Optimisation, Cybernetics, Algorithms

This book describes what was once thought not to be a feasible nor indeed viable exercise. This is to examine an applications program and search out any algorithms which may be considered "poor", and then replace them with algorithms that are programmed in a parallel form. There are many questions about how this can be achieved and the authors are at pains to explain their criteria. Most computer scientists would want a discussion about what is "poor",whether parallelizing the code rectifies the efficiency, and more important the choice of both strategy and method.

Should one write an entirely new applications program that employs not only parallelism to the application but any other approaches that offer "improvements"?The authors discuss this approach as well as one where the existing applications are converted to parallel form. The text's main descriptions are concerned with a discussion about a new theoretical framework for identifying "poor"algorithms and then replacing them with algorithms that parallelize the code. The book lives up to its title and concentrates on automatic recognition as well as replacement of algorithms. It is a new approach and readers should buy the book with an open and questioning mind.