GAME parallel support strategies for the parallelisation of genetic algorithms

The parallelism is introduced in the applications classified as time consuming ones to improve their performance measured generally by the execution time. Many studies and experiences have shown that the introduction of the parallelism in the genetic application improves the quality of the result as well as the execution time. This is a consequence of the optimisation technique used in the GA which is not deterministic one and of the role of the random in these algorithms. In fact, GAs offer a good deal of scope for parallelisation. Many of the typical GA operators (crossover, mutation and fitness evaluation) can be easily implemented in parallel, and would substantially benefit in running time if a parallel hardware platform is available. The parallelisation of an application consists in dividing the application into several tasks cooperating by message exchanging to fulfil the original application. According to the fitness of the decomposition we define the grain of the parallelism as the average size of the actions measured in the number of the executed instructions and the size of used memory. In the GA, the parallelism could be introduced at different levels: coarse-grain parallelism, intermediate granularity parallelism and fine grain parallelism.