Supercomputer Requirements for Selected Disciplines Important to Aerospace zyx VICTOR L. PETERSON, JOHN KIM, TERRY L. HOLST, GEORGE S. DEIWERT, DAVID M. COOPER, ANDREW B. WATSON, AND F. RON BAILEY zyxw Invited Paper Speed and memory requirements placed on supercomputers by five different disciplines important to aerospace are discussed and compared with the capabilities of various existing computers and those projected to be available before the end of this century. The disciplines chosen for consideration are turbulence physics, aero- dynamics, aerothermodynamics, chemistry, and human vision modeling. Example results for problems illustrative of those cur- rently being solved in each of the disciplines are presented and discussed. Limitations imposed on physical modeling and geo- metrical complexity by the need to obtain solutions in practical amounts of time are identified. Computational challenges for the future, for which either some or all of the current limitations are removed, are described. Meeting some of the challenges will require computer speeds in excess of exaflop/s zyxwvutsrq (10” flop/s) and memories in excess of petawords words). Finally, some evi- dence is presented to show that massively parallel computer archi- tectures, together with carefully constructed numerical methods optimized for these architectures offer the prospect of eventually meeting the computational power requirements for the selected disciplines. INTRODUCTION Computers have become indispensable in many of the aerospace science and engineering disciplines. One of the first disciplines to be treated effectively with a digital com- puter was flight mechanics. In fact, the ENIAC computer was developed during World War II to calculate ballistics firing tables for cannons. Today problems involving flight mechanics and orbital dynamics are routinely solved with- out the use of validating experiments. The design of aircraft structures is now done largely using computers, with only minimal proof testing of the finished product. Advances in fluid dynamicsand aerodynamics thatwould not have been possible without computers were first made about 20 years ago, when transonic solutions for a practical lifting airfoil with an embedded shock wave appeared in the literature [I]. Revolutionary advances in many other disciplines are now being made possible by the digital computer. The kinds of problems that can be solved effectively on Manuscript received September zyxwvutsrqpo 8, 1988; revised November 9, Theauthors arewith NASAAmes Research Center, Moffett Field, IEEE Log Number 8928434. 1988. CA 94035, USA. computers are strongly influenced by three factors: 1) the complexity of the underlying physics, 2) the availability of numerically stable and efficient algorithms for solving the governing equations on a given computational system, and 3) the amount of computational power (speed and memory) available. These factors determine the computer time required to obtain solutions. The use of 100 or more hours of computer system time to obtain a single solution may be practical in a research environment, but the effective use of computers to solve problems encountered in everyday applications requires that solutions be obtained in far shorter times, perhaps an hour or less. For the illustrative examples in this paper, practical amounts of computer time for research problems and applications problems have been chosen to be200 hours and 15 min, respectively. Of course, problems too complex to solve in reasonable amounts of time often can be simplified while retaining enough of the essential physics such that solutions obtained in shorter periods of time still provide valuable technical insights. Additional background on the importance of computers to some of the aerospace disciplines is given in [2], [3]. The purpose of this paper is to provide a brief overview of the requirements placed on supercomputers by selected aerospace disciplines. The subject is treated by examining some examples related to turbulence physics, aerody- nam ics, aerothermodynamics, chemistry, and human vision. In addition, some computer technologies needed to meet the future supercomputer requirements and their implications on numerical methods are discussed. The focus will be on future supercomputer hardware require- ments based on the use of current algorithms. Previous papers [2], [3] have shown, however, that algorithm research has resulted in increasingly more efficient numerical meth- ods. In fact, for the period from 1970 to 1983, reductions in solution times for the fluid dynamic equations due to improved algorithms were about the same as the reduc- tions resulting from faster machines. Prospects for even more efficient algorithms are very good for all of the dis- ciplines considered herein. Therefore, the current esti- mates of supercomputer requirementsare likelyto be upper bounds, particularly if expected increases in algorithm effi- ciency are realized. U.S. Government work not protected by U.S. copyrlght. 1038 PROCEEDINGS OF THE IEEE, VOL. 77, NO. 7, JULY 1989