• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2014, Vol. 36 ›› Issue (12): 2400-2409.

• 论文 • 上一篇    下一篇

一种新型混合互连网络拓扑结构的分析与优化

杨明英,雷斐,董德尊,沈胜宇,庞征斌   

  1. (国防科学技术大学计算机学院,湖南 长沙 410073)
  • 收稿日期:2014-09-15 修回日期:2014-11-14 出版日期:2014-12-25 发布日期:2014-12-25
  • 基金资助:

    国家863计划资助项目(2013AA014301,2013AA01A208);全国优秀博士学位论文作者专项资金资助项目(201450);国家自然科学基金资助项目(61272482,61303066)

Analysis and optimization of a new hybrid
interconnection network topology           

YANG Mingying,LEI Fei,DONG Dezun,SHEN Shengyun,PANG Zhengbin   

  1. (College of Computer,National University of Defense Technology,Changsha 410073,China)
  • Received:2014-09-15 Revised:2014-11-14 Online:2014-12-25 Published:2014-12-25

摘要:

随着高性能互连网络规模的增大,如何通过互连网络拓扑结构的设计来提升系统的性能和降低物理开销成为了系统设计的关键之一。传统的拓扑结构(可分为直接网络和间接网络)在网络规模增加时,不能很好地折衷网络性能和物理开销的关系。2012年Roberto P等人提出一种新型混合的拓扑结构,结合了直接网络和间接网络的特点,有效考虑了物理开销和网络性能的折衷。在此基础上,将新型混合拓扑每一维上的唯一的一个间接网络优化为多个间接网络,经过理论分析和实验模拟新型混合拓扑结构优化后的混合拓扑结构以及较常用的传统拓扑结构,优化后的混合拓扑结构能够在提升网络性能的同时降低物理开销。

关键词: 直接网络, 间接网络, 混合拓扑结构, 物理开销, 网络性能, 优化

Abstract:

With the expansion of highperformance interconnect network,how to improve the performance and reduce the physical overhead of the system through the design of interconnection network topology has become crucial in system design.The traditional topology,which can be divided into direct and indirect networks,does not keep a perfect balance between network performance and the physical overhead when the network size increases.In 2012,Roberto proposed a new family of hybrid topology,which combines the features of the direct network and indirect network and makes a balance between network performance and physical overhead. Based on these,the new family of topology can be optimized to be hybrid topology which owns more than one indirect networks.Theory analysis and simulation of both the new family of hybrid topology and the traditional topologies show that the optimized new family of hybrid topology can improve network performance while reducing physical overhead.    

Key words: direct network;indirect network;hybrid topology;physical overhead;network performance;improvement