In this article a parallel implementation of a quite recently introduced Differential Evolution algorithm for stochastic non-linear optimization is discussed. A new approach for efficient parallel implementation of Differential Evolution using a cluster of workstations connected via Local Area Network is suggested and the topics involved are discussed. This approach provides the required speed-up for optimization of computationally expensive objective functions such as computer simulation models of various technical systems.

Shared disk files are used for introducing an asynchronous communication channel between the master and slave processes. The use of disk files makes it possible to implement the program without any special programming tools, like PVM or MPI. Furthermore, no special hardware is required. For example the most widely available platform, a cluster of PCs connected via Ethernet, can be used.

Because the master process and slave processes are coupled only loosely via the shared interface files, the number of slave processes can be altered even during the optimization run. Both steady-state and generational reproduction of individuals can be used. Unlike than standard approach for parallelizing evolutionary optimization algorithms, the maximum number of involved slave processes is not limited by the population size of the master process.

The other major advantages of the suggested parallel computing approach are easy implementation, flexibility, robustness and low idle times of slave processes resulting in a high efficiency of parallelization.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £70 +P&P)