采用集簇方法的片上网络动态映射算法
A Dynamic Mapping Algorithm for Network？？on？？Chip Based on Cluster

针对片上网络(NoC)传统一一对应映射关系造成的资源节点利用率不高和通信功耗大等缺陷进行了改进,提出了一种采用集簇方法的NoC动态映射算法(DMA)。首先利用分枝界定算法完成通信量大且相连任务节点的集簇,减小了任务图通信总量;然后在此基础上借助自适应粒子群算法完成最优映射结果的获取;最后利用动态迁移策略对最优映射结果中单独占用资源节点的任务节点进行集簇。仿真实验表明,与随机映射、动态螺旋映射算法和最优邻居算法相比,DMA算法的通信功耗分别下降了73.93％、46.37％和14.55％,NoC面积占用率分别下降了50％、50％和33？？3％。
A dynamic mapping algorithm (DMA) for network？？on？？chip (NoC) based on cluster is proposed to solve the problems that the traditional one？？to？？one mapping relationship causes low utilization ratio of resource nodes and high communication power. Firstly, connected task nodes with high traffic load are clustered using a branch？？and？？bound approach to reduce the total communication volume of task graphs. Then an adaptive particle swarm algorithm is used to obtain the optimal mapping results. Finally, the dynamic migration strategy is used to cluster the task nodes that occupy an individual resource node in the optimal mapping results. Experimental results and comparisons with the random mapping, the dynamic spiral mapping algorithm (DSMA) and the nearest neighbor (NN) algorithm show that the proposed algorithm respectively reduces the communication power of the task graph by about 73.93％, 46.37％ and 14.55％, and the occupancy rate of the NoC area by about 50％, 50％ and 33.3％