A study of a non-linear optimization problem using a distributed genetic algorithm

Genetic algorithms have been used successfully as a global optimization method when the search space is very large. To characterize and analyze the performance of genetic algorithms on a cluster of workstations, a parallel version of the GENESIS 5.0 was developed using PVM 3.3. This version, called VMGENESIS, was used to study a nonlinear least-squares problem. Performance results show that linear speedups can be achieved if the basic distributed genetic algorithm is combined with a simple dynamic load-balancing mechanism. Results also show that the quality of search changes significantly with the number of processors involved in the computation and with the frequency of communication.