基于蚁群算法的负载优化方法研究

摘要

In allusion to the problem of high communication overhead between multi-core processors, this paper proposes a load optimization method based on ant colony algorithm. This method maps the interactive models to the same processor to reduce the communication of the simulation model. The problem of model communication is transformed into the model interaction problem. The computational load of the model is transformed into the model complexity problem. The ant colony algorithm is used to aggregate the simulation model with frequent interaction to reduce the communication cost. Considering the complexity of the model and the processing capability of each processor,the processor mapping of the simulation model is completed,so that the simulation model can optimize the system communication performance under the premise of calculating the load balance. And the ant colony algorithm corrects the shortcomings of the low initial pheromone which leads to the slow convergence of the algorithm. The simulation data show that this simplified method of"load optimization" can reasonably map the model to the processor and demonstrates the effectiveness of the method.%针对并行仿真在多核处理器间存在通信开销较大的问题,提出了基于蚁群算法的通信性能优化方法,该方法是将交互频繁的模型映射到相同的处理器上,以减少仿真模型的状态通信.将模型的通信问题转化为模型交互作用问题,将模型的计算负载转化为模型复杂度问题,通过蚁群算法将交互频繁的仿真模型进行聚合,以减少通信开销,根据聚合后的模型复杂度以及各处理器的处理能力,完成仿真模型的处理器映射,使仿真模型在计算负载平衡的前提下优化系统通信性能,并修正了蚁群算法初期信息素匮乏导致算法收敛速度慢的缺陷.仿真数据表明,通过这种简化"负载优化"的方法能够合理将模型映射到处理器上,并能较好的实现仿真的并行,效率的大幅度提高验证了该方法的有效性.