Performance Evaluation of DIRECT Algorithm in Parallel Optimization for a Thermohydrodynamic Lubrication Analysis

Abstract

In this study the optimization is illustrated by maximizing the load-carrying capacity of a slider bearing represented by a thermohydrodynamic lubrication model. An efficient optimization scheme, DIviding RECTangles (DIRECT), is coded for parallel computing in a single-system image cluster. The main focus of this study was to examine the performance of the standard DIRECT algorithm in coarse-grained parallelism. The parallel efficiency is compared with the results obtained in recent studies of air bearing optimization by using a genetic algorithm and a divide-and-conquer scheme. It was found that the limited possibility for parallelism of the standard DIRECT algorithm contributes somewhat small parallel efficiency.