Optimum design of externally pressurized air bearing using Cluster OpenMP

This study presents a performance evaluation of a new portable parallel programming paradigm, the Cluster OpenMP (CLOMP) for distributed computing, in conducting an optimum design of air bearings. The multi-objective optimization was carried out by using a genetic algorithm (GA) incorporating Pareto optimality criterion. Since the GA is natural parallel evolution algorithm, the computation of the search was carried out in parallel by using the CLOMP. In this study, the performance of a CLOMP cluster of four dual-core computers for the air bearing optimization was compared with a shared-memory processing (SMP) computer equipped with two quad-core processors. To examine the parallel efficiency of the CLOMP in the GA optimization, several multithread applications of various task sizes were tested. It is shown that the air bearing optimization can be effectively dealt with by the CLOMP (parallel efficiency of 96.2-98.8%) as well as the SMP computing (93.1-99.4%) in the studied cases. The CLOMP retains the characteristics of directive-based OpenMP, such as incremental programming and serial-coding compatibility. The verified high parallel efficiency of the CLOMP cluster demonstrates its potential applications of the scalable computing in many tribological optimizations.