一种片上网络缓存与带宽折中算法  

Algorithm of Buffer and Bandwidth Tradeoff in NoC

作　　者：樊利芳 张兴明[1] 董春雷 FAN Lifang;ZHANG Xingming;DONG Chunlei(Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学

出　　处：《信息工程大学学报》2018年第3期282-286,共5页Journal of Information Engineering University

基　　金：国家863计划资助项目(2009AA012201);河南省自然科学基金资助项目(08dz501600)

摘　　要：缓存和带宽是影响NoC性能的两个重要因素。大缓存可以承载更大的负载量,而大链路带宽允许缓存更快速的输入输出和更频繁的链路暂停,由此减小网络对缓存的需求。由于缓存和链路带宽争夺蕊片资源,需权衡两个因素之间的关系。针对缓存和带宽的折中算法进行了研究,得到缓存减小率和链路带宽增加率之间的关系。仿真结果表明,该折中算法提高了片上网络的整体性能,尤其使面积开销减小了16. 9%。Buffer and bandwidth are two important factors that influence the performance of NoC. Large buffers allow larger workload absorbing. However, higher bandwidth allows faster buffer draining and more link pausing, which reduces buffering demand for network. As the two features compete for silicon resources, modeling their relative impact on the network is critical. So the tradeoff between buffer and bandwidth is studied to obtain the relationship between buffer reducing rate and bandwidth increasing rate. The simulation result show that the algorithm improves the overall performance of the network, especially the area is reduced by 16.9%.

关 键 词：NOC 缓存 带宽 折中 

分 类 号：TP3-05[自动化与计算机技术—计算机科学与技术]