Computer system considerations for large-scale optimization projects

摘要

For years now vector supercomputers have provided the only means by which large, full-scale automotive body structures can be analyzed for stress, strain, deflection and vibration response within an acceptable amount of time. RISC-based supercomputer systems are now available, challenging the pre-eminence of traditional vector supercomputers at much less cost. At the same time, trends in Engineering analysis are towards larter, more detailed models, longer simulation times and more sophisticated analyses, increasing the computer time required for a solution. At the forefront of this trend is structural optimization - a method by which not just a single solution is obtained for design verification, but a much more complete set of solutions is computed and an overall 'best' design is found. But as the set of input parameters expands the solutions require increased computer resources at a non-linear rate. Structural optimization is able to provide insight into a better design; less mass, higher performance, etc. It is most useful to the overall design process if it is employed earlier than traditional verification analyses and so comes with a higher time, design cycle and scheduling premium. To solve such a large-scale problem in an acceptable time a compute 'engine' of multiple cpus and machines and networks needs to be utilized. This paper will discuss some of the computer resource issues involved with structural optimization and how they could be tackled in today's large-scale corporate environment.