An Asynchronous Parallel Differential Evolution Algorithm

A Parallel Differential Evolution algorithm is presented in this work, developed for a cluster of computers in Windows environment. The parallelization is realized using an asynchronous approach, utilizing a Master-Slave architecture. A separate executable program is used to evolve each member of the population. The current population is stored in a folder accessible by all executables; each current member of the population, along with its fitness, is stored in a separate text file contained in this common folder. Each slave program uses the information stored in the common folder to evolve the corresponding member of the population and to update the information stored in the corresponding text file, independently from the rest executables. More than one executables may be assigned to each computer. The procedure is tested in two airfoil optimization problems and the parallel code is compared to a serial one, with respect to the convergence behaviour, the quality of the optimum solution and the total computation time.