Genetic Algorithms (GAs) are being used to solve a wide range of problems in real world problems, and it is important to study their implementations to improve the solution quality and reduce the execution time. Designing parallel (e.g., distributed) GAs is one research line to do so. In distributed GAs, every individual represents a tentative solution. Individuals are split (and sparsely communicated) over many islands, with genetic operators being applied locally in each island. In addition, in order to maintain diversity and reduce the number of the evaluations, a migration operator is used to enhance their behavior. This article presents a basic study on the speed-up of parallel GAs where a common approach is followed to better understand synchronous and asynchronous versions together. We analyze the behavior of GAs over a homogeneous multiprocessor system. We will report results showing linear and even super-linear speed-up in both cases of study. The parallel performance of the synchronous and asynchronous versions is very good in a multiprocessor computer, both in terms of time and solution quality. Besides, a statistical analysis of the algorithms clearly proves that both cases have a similar numerical behavior over a homogeneous parallel system.
