Data Communication Requirements for the Advanced Nas Network

The goal of N A S A ' s ,Vumerical A4erodynamic Simulation (.VAS) Program is to provide a powerful computational environment for advanced research and development in aeronautics and related disciplines. The present .VAS system consists of a Cray 2 supercomputer connected b y a data network t o a large mass siorage system, t o sophisticated local graphics workstations and b y remote communications to researchers throughout the United States. The program plan is to continue acquiring the most powerful supercomputers as they become available. I n the 1987/1988 time period it is anticipated that a computer with 4 times the processing speed o f a Cray 2 will be obtained and by 1990 a n additional supercomputer with 16 t imes the speed of the Cray 2. This paper describes the implications of this 20-fold increase in processing power on the data communications requirements. The analysis was based on models o f the projected workload and sys fem architecture. The results are presented together with estimates of their sensitivity to assumptions inherent in the models. *This work wasTupported in part by Cooperative Agreement NCC 2-387 from the National Aeronautics and Space Administration ( S A J A ) to the Universities Space Research Association (USRA). DATA COMMUNICATION REQUIREMENTS FOR THE ADVANCED NAS NETWORK INTRODUCTION: The Numerical Aerodynamic Simulation (NAS) Program was initiated by NASA to establish a national resource for advanced research and development in aeronautics and related disciplines. To achieve this goal the NAS Program is to, "act as the pathfinder in advanced, large-scale computer system capability through systematic incorporation of state-of-the-art improvements in computer hardware and software technologies". The justification, historical background, technical objectives and long term plans of the NAS Program have been presented previously [ 1, 2, 31. The first major milestone of the NAS Program has now been achieved and the initial operating configuration of the NAS Processing System Network (NPSN) located at the NASA Ames Research Center in California is shown schematically in Figure 1. (See [4] for detailed specifications). The centerpiece is a Cray 2 supercomputer with 250 million (64-bit) words of memory and a sustained performance rated a t 250 MFLOPS (250 Million FLoating-point Operations per Second) as measured on a set of S A S benchmark tests for optimized large-scale computational aerodynamic application codes. This impressive capability is enabling researchers in the aerophysics field to address previously unsolved problems and to gain insight into complex aerodynamic phenomena. However it is only the beginning. In recognition of the on-going nature of the S A S program, the Cray 2 is designated HSP-1 (High Speed Processor 1). The need for much more powerful processors can be seen from Figure 2 (1, 2, 31 which depicts the estimated speed and memory requirements for various levels of approximation to the governing fluid-dynamics equations for three levels of geometric complexity; an airfoil, a wing and a complete aircraft. Note, for example, that if viscosity effects are included by using the Reynolds-averaged SavierStokes equations, a three dimensional solution for a wing reqiires about 100 times the computing speed of a comparable inviscid solution and only now with the Cray 2 is it feasible to perform highly repetitive design optimization studies for such cases. Furthermore, if still more realistic large eddy effects are to be considered, a further factor of about 1000 is required for runs cf the same duration. Finally, Figure 2 indicates that (with 1985 algorithms). a l-ingle 15 minute run including large eddy effects for a complete aircraft would reljuire computer speed in excess of 10" floating point operations per second and tin estimated random access memory of 10" bytes! In consideration of these computational needs, the NAS Program plan is to continue to acquire the most powerful supercomputers as they become available. In the 1987/1988 time period it is anticipated that a "one GigaFLOPS" computer (HSP-2) with 4 times the speed of a Cray 2 will be obtained and by 1990 an