Parallelization of a minimization problem for multi-processor systems

Abstract

A simulation and analysis program system for the design and the analysis of catalytic fixed-bed reactors was selected for parallel implementation. The minimization subroutine of this program system was adapted for parallel computation both at the algorithm level and at the program level. The algorithm follows the Simplex method of Nelder and Mead and allows optimization of the reaction parameters minimizing a least-squares-function. The computation of this minimization problem was prepared for the Erlan-gen-General-Purpose-Array (EGPA) multiprocessor-system.

First results obtained with test-functions showed a strong dependence of the utilization of the multiprocessor system as a function of computation time. In the most favourable case, where a test-function with long computation time was minimized a speedup of 3.7 (multiprocessor system compared to a monoprocessor; theoretical upper bound 5.0) was achieved.

This work was sponsered by the Federal Ministry for Research and Technology, Bonn (contract no. :DV 4.906 - 081 2070 A).