Optimal Many-to-Many Personalized Concurrent Communication in RapidIO-based Fat-trees

	Optimal Many-to-Many Personalized Concurrent Communication in RapidIO-based Fat-trees
	Shu Lin; Hao,Jie; Song,Yafang; Li,Chengcheng; Wang,Donglin; Shu,Lin
	2016-05
会议名称	17th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing (SNPD 2016)
会议录名称	17th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing (SNPD 2016)
会议日期	2016.05.30-2016.06.01
会议地点	Shanghai,China
摘要	Many-to-many (M-to-N) personalized concurrent communication is one of the most dense traffic patterns in high performance computing system. Each node in group M sends different messages to every node in group N. Hot-spot congestions are the bottleneck of this communication pattern. In this paper, a proactive congestion avoidance mechanism is proposed to optimize many-to-many personalized concurrent traffic in RapidIO-based fat-trees. And a heuristic scheme of virtual destination is proposed to deal with the case that messages from each node in M group to every other node in N group are not of the same size. Our proactive congestion avoidance mechanism has been used in a practical network, which is a part of digital signal processing platform for “the TianLai Pathfinder”. The network delay is effectively improved, reducing from about 34.784 us to 1.484 us. And the communication overhead is decreased by twice.
关键词	Many-to-many Personalized Rapidio Fat-tree Node-level Congestion-avoidance
收录类别	EI
文献类型	会议论文
条目标识符	http://ir.ia.ac.cn/handle/173211/11730
专题	国家专用集成电路设计工程技术研究中心
通讯作者	Shu,Lin
作者单位	Institute of Automation, Chinese Academy of Sciences
推荐引用方式 GB/T 7714	Shu Lin,Hao,Jie,Song,Yafang,et al. Optimal Many-to-Many Personalized Concurrent Communication in RapidIO-based Fat-trees[C],2016.

条目包含的文件		下载所有文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
Optimal Many-to-Many（775KB）	会议论文		开放获取	CC BY-NC-SA	浏览下载