Performance evaluation of parallel genetic algorithms for optimization problems of different complexity

Publisher Summary

This chapter discusses performance evaluation of parallel genetic algorithms for optimization. In recent years, Genetic Algorithms (GAs) have proved to be very efficient in dealing with a large variety of optimization problems. Yet their needs of execution time are very high especially for simulation optimization, which makes parallelization desirable. Genetic Algorithms are the most widely examined subgroup of EAs. They operate on a set (population) of possible solutions (individuals) of a problem. An individual is an element of the space of solutions spanned by the problem-relevant parameters. The fitness of an individual describes its quality to solve the corresponding problem. The GA propagates the members of a population by genetic operators, simulating biological procedures (selection, mutation, and recombination) to improve the fitness of the individuals and to find the individual optimally solving the problem. This chapter discusses three different parallel implementations of a Genetic Algorithm: a global master-slave, a coarse-grained and a hierarchical parallel GA (PGA). Their performance is evaluated by applying them to several optimization tasks and performing runtime tests on a large Beowulf linux cluster.