Aiaa - 92 - 2734 - Cp Future Directions in Computing and Cfd

generations of computers as well as improvements in algorithms. As a results CFD has become an ever more powerful tool in the design of aerospace systems. The high computational demands of aeroscience applications have been the driving force behind the recent rapid development of CFD. Moreover, CFD has been an equally successful modeling tool in a variety of other fields, such as automotive engineering. Now, in the 1990's it has become apparent that conventional supercomputers cannot sustain CFD's continued high rate of advancement. Thus CFD has reached a critical juncture, since it is becoming more and more apparent that future growth in computational speed will be increasingly the result of parallel processing technology. In recent years CFD on massively parallel machines has become a reality. In this paper we summarize some recent trends both in high performance computing, and in CFD using parallel machines. We discuss the long term computational requirements for accomplishing some of the large scale problems in computational aerosciences, and current hardware and architecture trends. We present performance results obtained from the implementation of some CFD applications on the Connection Machine CM2 and the Intel iPSC/860 at NASA Ames Research Center. Finally we argue that only massively parallel machines will be able to meet these grand challenge requirements, and we outline the computer science and algorithm research challenges ahead. In the last several years a wide variety of parallel machines have become available for exploring the issues of using parallelism in scientific computing in general and CFD in particular. Most of the early ("zero-th generation") machines that appeared between 1983 and 1987 were rather experimental in nature, and served mainly for research investigations in areas such as algorithms, languages, and operating systems for parallel computing. In 1988 and 1989 several members of a first generation of parallel supercomputers became available. We want to use the term "supercomputer" here because these parallel supercomputers such as the Connection Machine CM-2, the Intel iPSCj860, the NCUBE2, and others are in their larger configurations are comparable both in memory and peak computational speed to the performance of the most powerful conventional supercomputers, e.g., the Cray Y-MP. However, it is well