The applications on distribution logistics are diverse, such as the transportation planning and delivery of goods or in telecommunication networks data routing. Given the breadth and capillarity of these problems, studies have been developed to reduce network operating costs of this magnitude, especially regarding the demand for electricity. Therefore, this work proposes a method of resolution of routing problems with high demand. The proposed method is based on bio-inspired algorithms, which combined with other methods, ensure the integrity of the solutions, as well as its proximity to optimum. Nevertheless, such algorithms becomes computationally expensive as the application complexity in question grows and, therefore, multiprocessor environment, like GPU Computing platforms, has being widely used to increase bio-inspired algorithms performance. Thus, this work aims perform tests about the widespread parallelization techniques of these algorithms, intending to make an evaluation of which strategies has better relation with each tested algorithm. In order to do this, the routing problem in WDW optics networks with high demand level was used as a case study, in which it is needed define which are the better routes to demands sent simultaneously. The algorithms that assisted the tests were Genetic Algorithms and Swarm Particle Optimization, which are highly disseminated. The results show that the parallelization strategy to be used depends as much on the platform in which has been implemented, as the problem to be treaty.
|
