A high-efficient multi-deme genetic algorithm with better load-balance

Genetic algorithm is a very powerful search algorithm that fits for many complex situations. However, it is very time consuming, which limits its usage. Previous work which makes use of multi-core systems to parallelise it performs well and gains much attention. This paper introduces that the load-imbalance problem in parallel genetic algorithm will incur large overhead and will limit the performance. We propose two efficient mechanisms (postponed waiting and work stealing) to achieve fine-grained schedule to solve the problem. Compared with traditional multi-deme parallel genetic algorithm, our high-efficient multi-deme genetic algorithm (HMGA) can achieve an average speedup of 1.36.