Concurrent Structural Optimization on Massively Parallel Supercomputer

Abstract

Genetic-algorithm (GA)–based structural optimization can be parallelized to a high degree on new generation of scalable distributed-memory multiprocessors. In this paper, a mixed computational model is presented for GA-based structural optimization of large space structures on massively parallel supercomputers. Parallelism is exploited at both coarse-grained design optimization level in genetic search using the multiple-instruction–multiple-data model of computing and fine-grained fitness function evaluation level using the single-instruction–multiple-data model of computing. The latter model involves the development of a data-parallel iterative preconditioner-conjugate-gradient algorithm for the solution of the resulting system of linear equations. The model has been implemented on Connection Machine CM-5 and applied to optimization of large space steel structures subjected to the constraints of the American Institute of Steel Construction's allowable stress design specifications. The model and concurrent algorithm developed in this research is highly scalable. A peak performance of 2.4 giga–floating-point operations per second is achieved using 512 processors for a structure consisting of 4,016 elements.