AIAA 96–0409 Automatic Aerodynamic Optimization on Distributed Memory Architectures

This paper presents a parallel implementation of an automatic Euler design method based on the control theory of systems governed by partial differential equations. The Euler equations and the resulting adjoint equations necessary to calculate the Frechet derivatives for the gradient of the cost function are solved using a domain decomposition approach with communication handled by the MPI (Message Passing Interface) Standard. Parallel performance is evaluated on a distributed memory parallel computer and sample calculations are presented. A complete optimization procedure on a 192 × 32 × 48 mesh can be completed in 7 minutes using 16 processors of an IBM SP2 system. This clearly shows that parallel processing is a key enabling technology for CFD to become an efficient tool in a realistic design environment. The parallel implementation of a multiblock version of the program which allows for a higher degree of geometric complexity in the design has recently been completed. Parallel performance trends of the multiblock code are consistent with the ones observed in the single block implementation.