Supercomputing in Engineering Analysis NEW GENERATION COMPUTING Series Editor Hojjat Adeli Department of Civil Engineering The Ohio State University Columbus, Ohio 1. Super-computing in Engineering Analysis, edited by Hojjat Adeli 2. Parallel Processing in Computational Mechanics, edited by Hojjat Adeli Supercomputing in Engineering Analysis edited by Hojjat Adeli The Ohio State University Columbus, Ohio Marcel Dekker, Inc. New York • Basel • Hong Kong Library of Congress Cataloging-in-Publication Data Supercomputing in engineering analysis / edited by Hojjat Adeli. p. cm. -- (New generation computing; 1) Includes bibliographical references and index. ISBN 0-8247-8559-2 (alk. paper) 1. Engineeering-Data processing. 2. Supercomputers. 3. Vector processing (Computer science) 4. Parallel processing (Electronic computers) I. Adeli, Hojjat. II. Series. TA345.S87 1991 620'.00285'41 l-dc20 91-25607 CIP This book is printed on acid-free paper. Copyright © 1992 by MARCEL DEKKER, INC. All Rights Reserved Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, microfilming, and recording, or by any information storage and retrieval system, without per- mission in writing from the publisher. MARCEL DEKKER, INC. 270 Madison Avenue, New York, New York 10016 Current printing (last digit): 10 9 8 7 6 5 4 3 21 PRINTED IN THE UNITED STATES OF AMERICA Preface This is the first volume in the series New Generation Computing. Contributed by leading researchers, the volumes in the series are aimed at the most recent advances in computer science and new and emerging computer technologies. This volume as well as the second volume of the series, Parallel Processing in Computational Mechanics, concentrate on high-performance computing using the new generation of computers with vector and parallel processing capabilities. These machines have provided the means to tackle problems not readily tractable on traditional computers. The volumes are introductory in nature and are intend- ed primarily for an engineering audience. They are also valuable for computer scientists and software developers. The first six chapters of this volume present general aspects of supercom- puting from both hardware and software engineering points of view. In the first chapter parallel computing in distributed computing systems is presented as an effective way of building a wide range of supercomputing applications and providing a flexible user environment for scientists and engineers. Chapter 2 presents a thorough comparative description of advanced architecture computers that are available commercially, with an emphasis on architectures that use some form of parallelism. Various VLSI architectures for engineering supercomputing are reviewed in Chapter 3. The topic of benchmarking supercomputers and minisupercomputers is discussed in Chapter 4. An overview of principles and techniques for automatic vectorization and parallelization of numerical FOR- iii iv PREFACE TRAN programs on both shared-memory and distributed-memory architectures is given in Chapter 5. These techniques can be applied to a wide range of engineering problems. Chapter 6 presents supercomputing with data-driven wavefront array processors, with an emphasis on the direct hardware implemen- tation of a matrix procedure for solution of a linear system of equations, which is ubiquitous in engineering disciplines. This chapter concludes with remarks on the simulation of the phenomenon of the propagation of the waves. Chapter 7 presents homotopy algorithms, which are particularly suitable for highly nonlinear problems for which initial solution estimates are not readily available. Such nonlinear systems of equations are encountered in numerous engineering problems. Homotopy algorithms are inherently parallel to a great extent and therefore suitable for implementation on parallel architectures. Commercial finite-element software packages with thousands of subroutines are used extensively for analysis of many complex engineering problems. Ac- curate modeling of these problems with tens of thousands of degrees of freedom may require an inordinate amount of computational resources. Consequently, efficient implementation of finite-element programs is an important problem, which is discussed in Chapter 8. Fluid mechanics has important applications in several engineering disci- plines: mechanical engineering, aeronautical engineering, and civil engineering. Parallel computations of complex engineering flows are described in Chapter 9. Chapter 10 presents modeling of multiphase, multicomponent fluid flow through porous media and reservoir simulation on a supercomputer. The topics of supercomputing and parallel processing are akin to each other. Therefore, the volume Parallel Processing in Computational Mechanics, which covers such topics as finite element analysis on supercomputers and parallel solution of fluid mechanics problems, should be considered an accompanying volume to this book. Hojjat Adeli Contents Preface Hi Contributors vii About the Editor xi 1. Supercomputing in the 1990s: A Distributed Solution 1 S. Ashley Burns and Charlie F. Bender 2. Advanced Architecture Computers 19 Jack Dongarra and Iain S. Duff 3. VLSI Architectures for Supercomputing 63 Tse-yun Feng and Chuan-lin Wu 4. Benchmarking Super- and Minisupercomputers 111 Wolfgang Gentzsch 5. Automatic Restructuring for Parallel and Vector Computers 135 Hans Zima and Barbara Chapman 6. Supercomputing with Data-Driven Wavefront Array Processors 169 David John Evans and Michael P. Bekakos vi CONTENTS 7. Homotopy Algorithms for Engineering Analysis 199 Layne T. Watson and Manohar P. Kamat 8. Optimization of Finite-Element Software Systems on Supercomputers 261 Louis Komzsik 9. Parallel Spectral Computations of Complex Engineering Flows 289 George Em Karniadakis and Steven A. Orszag 10. Supercomputing in Reservoir Simulation 325 Richard E. Ewing Index 361 Contributors Michael P. Bekakos received his Ph.D. from the Parallel Algorithms Re- search Centre of the Loughborough University of Technology, Loughborough, England. He is currently an assistant professor in the Department of Applied Informatics at the Athens University of Economics and Business, Athens, Greece. He has published papers on parallel processing and systolic algorithms. Charlie F. Bender received his Ph.D. from University of Washington in 1968. He is currently Director of the Ohio Supercomputer Center and Director of the Office of Research Computing at The Ohio State University. His previous ap- pointments include Director of the Advanced Computational Methods Center at the University of Georgia and Head of the Chemistry and Materials Science Department at Lawrence Livermore Laboratory. He is the author of over 115 research and technical publications. S. Ashley Burns has written about large-scale computing for a variety of technical publications. She is the Manager of Information Systems at the Ohio Supercomputer Center. Barbara Chapman is on the faculty of the Computer Science Department at the University of Vienna, Vienna, Austria, where she teaches courses in compiler construction, parallel systems, and programming methods for parallel computers. She is the co-author of the book Supercompilers for Parallel and Vector Computers. Her research interests are in algorithms and programming vii