We have designed and implemented a parallel version of the Generalised Reduced Gradient optimisation method (GRG), especially devised for efficient processing on heterogeneous NOWs. The core parallel routines deal with simultaneous constraint evaluation and the calculation of gradients for both the objective function and the constraints. A hybrid model for task scheduling that minimises idle time and considers the heterogeneous nature of the processors was proposed. As to performance comparisons, a modified speed-up metric that takes into account heterogeneity was employed. Significant time improvements were obtained for both academic and industrial examples corresponding to process-plant units. The best results were attained for large-scale problems or when the functions to be evaluated were costly.
展开▼
