电子学报 ›› 2014, Vol. 42 ›› Issue (11): 2142-2149.DOI: 10.3969/j.issn.0372-2112.2014.11.003

• 学术论文 • 上一篇    下一篇

一种故障通道隔离的低开销容错路由器设计

欧阳一鸣1, 陈义军1, 梁华国2, 易茂祥2, 李建华1   

  1. 1. 合肥工业大学计算机与信息学院, 安徽合肥 230009;
    2. 合肥工业大学电子科学与应用物理学院, 安徽合肥 230009
  • 收稿日期:2013-10-16 修回日期:2014-01-20 出版日期:2014-11-25
    • 通讯作者:
    • 欧阳一鸣
    • 作者简介:
    • 陈义军 男,1990年生,硕士研究生,研究方向:片上系统以及片上网络容错方法. E-mail:yijunchen_hf@163.com;梁华国 男,1959年生,教授,博士生导师,中国计算机学会容错计算专业委员会委员,研究方向:嵌入式系统综合与测试、数字系统设计自动化、ATPG算法与分布式控制等. E-mail:huagulg@hfut.edu.cn
    • 基金资助:
    • 国家自然科学基金 (No.61474036,No.61371025,No.61274036,No.61106038)

Design of a Low-Overhead Fault Channel Isolated Fault-Tolerant Router

OUYANG Yi-ming1, CHEN Yi-jun1, LIANG Hua-guo2, YI Mao-xiang2, LI Jian-hua1   

  1. 1. School of Computer and Information, Hefei University of Technology, Hefei, Anhui 230009, China;
    2. School of Electronic Science & Applied Physics, Hefei University of Technology, Hefei, Anhui 230009, China
  • Received:2013-10-16 Revised:2014-01-20 Online:2014-11-25 Published:2014-11-25

摘要:

片上网络中路由器发生故障势必会影响整个网络的性能,过大的容错开销也会给网络带来很大的负担.对此,本文提出了一种故障通道隔离的低开销容错路由器架构,该路由器通过减少不必要的交叉开关及合理优化各个端口VC的数目来减小路由器整体开销,同时增加一个冗余通道来达到对路由器容错的目的.当路由器中某个通道发生故障时,通道隔离检测方法使路由器能够在检测故障类型的同时进行数据传输,带回收指针的重传buffer将会进一步减少整个容错结构的开销.实验结果表明在无故障情况下本文设计的路由器较传统路由器平均延时降低45%左右,最大吞吐率提高28%左右,面积开销仅仅增加了18.24%.在故障存在的情况下,本文方案也显现出很大的优越性,能够达到很好的容错效果.

关键词: 片上网络, 路由器故障, 容错, 故障通道隔离

Abstract:

A fault occurring in the router of Network-on-chip will certainly affect the overall network's performance.At the same time,the excessive fault-tolerant overhead brings a huge burden to the network.For these reasons,this paper proposes a low-overhead fault-tolerant router architecture which reduces the area overhead by decreasing unnecessary crossbar and rationally optimizing the number of VCs in each port.Simultaneously,a redundant channel is used to tolerate fault for the router.When a channel fault occurs in the router,by using an isolated channel detection method,the routers can simultaneously detect the fault type and transmit data.Moreover,a new buffer with a recycle pointer is designed to retransmit error packages which would save the overhead of the fault-tolerant unit.Experimental results show that,in the case of no fault existing,the proposed router has 45% lower latency in average and 28% higher throughput rate performance than a typical VC router with only 18.24% larger area.In the case of the faults existing,the proposed router is also able to achieve good fault tolerance effect.

Key words: network-on-chip, router fault, fault-tolerant, fault channel isolated

中图分类号: