胖树型片上网络的低能耗映射算法

A Low-Energy Mapping Optimization Algorithm for Fat-tree Based Network-on-Chip

为解决胖树型片上网络的映射问题,针对该网络不同层路由器间链路长度不同的特点,提出一种低能耗映射优化模型,并设计一种基于捕食搜索策略的差分进化算法(PSDE)进行模型求解.该算法将捕食搜索策略与差分进化算法相结合,弥补了差分进化算法易陷入局部最优的不足,增强了捕食搜索策略的局部搜索能力.仿真实验结果表明,与遗传算法和模拟退火算法相比,PSDE可以缩短运行时间,并获得低能耗、高质量的优化映射结果.

This paper copes with the problem of mapping for fat-tree based network-on-chip. Considering the difference of the link length between routers in different layers of the network, an energy-aware mapping optimization model is proposed. Differential evolution algorithm based on Predatory Search strategy （PSDE） is designed to solve the mapping model. Differential evolution algorithm is easy to fall into local optimum. To make up for this deficiency, PSDE is proposed to strengthen the concentrated local search ability of predatory search strategy. The experimental results show that, compared with genetic algorithm and simulated annealing algorithm, the PSDE can obtain lower energy mapping results and save running time.