Highly Scalable Parallel Genetic Algorithm on Sunway Many-Core Processors
Zhiyong Xiao,Xu Liu,Jingheng Xu,Qingxiao Sun,Lin Gan
DOI: https://doi.org/10.1016/j.future.2020.08.028
IF: 7.307
2020-01-01
Future Generation Computer Systems
Abstract:As a heuristic method, the genetic algorithm provides promising solutions with impressive performance benefits for large-scale problems. In this study, we propose a highly scalable hybrid parallel genetic algorithm (HPGA) based on Sunway TaihuLight Supercomputer. First, the Cellular model is presented on a thread level, so that each individual can be processed by a single computing unit which is in charge of the parallel fitness calculation, crossover, and mutation operations. The information exchange between individuals is realized by register communication. Second, the Island model is assigned to a process level, so that each process accounts for a single sub-population, and the migration among sub-populations is implemented using MPI communication. The proposed approach can fully exploit the individual diversity of the genetic algorithm and reasonably maintain the communication overhead. Based on the widely used CEC2013 benchmark, the experimental results show that the algorithm presents a sound performance in terms of both accuracy and convergence speed.