Parallel approach for genetic algorithm to solve the Asymmetric Traveling Salesman Problems
Yassine Moumen, Otman Abdoun, Ali Daanoun · 2017
Metaheuristics methods are the most efficient if not the only way to solve difficult problems in both science and industry. In computer science, these methods are being used to find a good answer for NP-hard optimization problems in moderate execution times by shrinking the size of the search space to focus on regions with the height change of having an acceptable solution. Nevertheless, if we consider a large problem instance as in the real word, finding a good solution with traditional implementation of metaheuristics needs a huge computational power (in term of processing capability and memory usage) as well as time to solve, even the best known machines in our time cannot handle the massive work load to just initials a real word scenario. For that reason, implementing a parallel computing of these methods is number one priority to speed up the search giving that in most cases, the biggest limitation is the time; one of the newest techniques to achieve the best results is by using Graphical processing units (GPUs). However, taking advantage of GPU's parallel nature to compute Metaheuristics is rarely studied in the literature. In this paper, we present a new approach for the design and implementation of effective metaheuristics algorithms on GPU by using the latest technologies like CUDA. To accelerate the search mechanism even more, we have introduce new functions like host-device data transfer optimization, thread control, Coalesced memory access.