NoC架构片上多处理器系统性能探索

Performance Exploring of NoC Based Multi-Processor System-on-a-Chip

采用SystemC建模和仿真环境建立了一款NoC系统级仿真平台,设计了3个实验分别用于建模3种典型应用(低计算/通信比、高计算/通信比和非独立任务),以定量模拟的方法对NoC架构MPSoC性能进行了详细的调研,并将其结果与总线架构MPSoC进行了对比分析.实验结果显示:NoC系统加速比与处理器数目呈线性关系,不受规模的影响,而总线系统则明显受到处理器数目的限制;共享存储资源成为NoC系统性能提升的限制,但可以通过采用分布式存储策略得到解决,而总线系统却无法克服其共享总线通信瓶颈.因此,在系统规模较大(N>12)时推荐采用NoC体系结构.

In this paper, a NoC system level simulation platform is setup first, then three kinds of tasks are designed to run on this platform and experimental results are analyzed and compared with that of bus-based MPSoC; lastly. It is shown that the speedup of NoC system is in linear with the number of integrated processors, which proves performance of NoC system is not restricted by processors number as bus-based system does; shared memory resources have important influences on NoC; system performance, but it will not become a bottleneck by utilizing distributed shared memory solution. So NoC is recommended when the size of target system is large （N〉12）.