A scalable algorithm for the parallel solution of simulation-based optimization problems

A common approach to the design and implementation of parallel optimization algorithms is the a posteriori parallelization of existing sequential algorithms. According to Amdahl's law, the theoretically achievable gain in performance using this type of "parallel" algorithm is limited by its sequential components. The paper with the design and analysis of an inherently parallel algorithm for the distributed solution of simulation-based nonlinear constrained optimization problems in engineering on a network of workstations. The characteristics of these nonlinear optimization and control problems in engineering lead to very time consuming solutions. The design of the algorithm and a quantitative analysis of its parallel performance is discussed. Results of its application to a mathematical test case and an industrial application from water engineering are used to show the feasibility of the proposed algorithm.