摘要
多核系统集成度不断增加,造成片上网络的延迟和吞吐性能迅速下降。提出基于参数分簇的层次化片上网络结构,设计3类基本分簇作为单元模块,搭建不同节点规模下的层次化结构。实验结果表明,在局部化流量模式下,文中结构比传统Mesh和Clustered Hierarchical Mesh在256节点时的延迟分别降低18.76%和17.31%,吞吐提升18.57%和12.97%,资源仅增加25%和10%。
With the increasing size of chip-multi-processor (CMP) , the latency and throughput per- formance of conventional NoC is getting worse. This paper proposes a new parameter-based hierar- chical mesh NoC (PHNoC). Three base-clusters are designed to form multi-layers and three param- eters are presented to design different size of CMP. Experimental results demonstrate that, compared with Mesh and CHMesh, PHNoC reduces latency by 18.76% and 17.31% and improves throughput by 18.57% and 12.97% , with only 25% and 10% more resources overhead respectively under size of 256 and localized traffic.
出处
《信息工程大学学报》
2016年第3期346-352,共7页
Journal of Information Engineering University
基金
河南省自然科学基金资助项目(122300413201)
关键词
片上网络
层次化互连
分簇结构
参数化设计
network-on-chip
hierarchical interconnection
clustered structure
parameterized design