面向时延的NoC映射技术研究

针对NoC任务映射问题中时延难以预测和启发式算法效率低的问题,提出一个时延改进模型和近邻随机遗传算法。该模型从宏观的链路负载分布和单个节点的排队时延两方面来构建NoC映射的时延模型,通过引入时延因子、权重系数来刻画不同映射方案对时延性能的影响,避免了NoC通信时延精确建模的难题。提出近邻随机思想来构建遗传算法的初始种群,并且运用该算法实现了面向时延的NoC映射,在达到全局最优的情况下,比经典遗传算法效率提升将近20%。实验结果表明,该算法优于现有的经典遗传算法和随机映射方案。

Research on delay-aware NoC mapping method

Due to the delay in NoC task mapping is difficult to predict,and the low efficiency in heuristic algorithm.This paper proposed an improved delay model and nearest-neighbor random genetic algorithm(NNRGA).It constructed the NoC mapping delay model from the macroscopic link load distribution and the single node queue latency.Different mapping schemes influenced the performance of delay through importing delay factor and weight coefficient,the model avoided the difficulty to model communicating delay in NoC accurately.This paper proposed a method to construct initial population of genetic algorithm based on the thought of nearest neighbor and random.It used NNRGA to realize the delay-aware NoC mapping.And the efficiency increased by nearly 20% compared with the classical genetic algorithm when achieving the global optimum situation.the experimental results show that the algorithm is better than the classical genetic algorithm and random mapping algorithm.