广东工业大学学报 ›› 2017, Vol. 34 ›› Issue (04): 97-101.doi: 10.12052/gdutxb.160020

• 综合研究 • 上一篇    下一篇

一种提高无缓存片上网络性能的方法

张坤, 刘怡俊   

  1. 广东工业大学 计算机学院, 广东 广州 510006
  • 收稿日期:2016-02-29 出版日期:2017-07-09 发布日期:2017-07-09
  • 通信作者: 刘怡俊(1977–),男,教授,博士,主要研究方向为计算机体系结构、集成电路设计.E-mail:yjliu@gdut.edu.cn E-mail:yjliu@gdut.edu.cn
  • 作者简介:张坤(1993–),男,硕士研究生,主要研究方向为片上网络.
  • 基金资助:

    广东省科技计划项目(2015B090908001,2013A090100005,2014B090901061,2015B090903080);广州市科技计划项目(2014Y2-00211)

A Method of Raising Bufferless Network-on-Chip Performance

Zhang Kun, Liu Yi-jun   

  1. School of Computes, Guangdong University of Technology, Guangzhou 510006, China
  • Received:2016-02-29 Online:2017-07-09 Published:2017-07-09

摘要:

无缓存片上网络路由在低负荷时表现较有缓存路由有能耗低、延迟低的优势,而在较高负荷情况下由于竞争的增加和较多的微片偏转的出现降低了无缓存路由的性能,使得延迟增加、能耗升高.本文从拥塞控制出发,提出了一种新的流控方式降低在高负荷情况下因为过多的微片偏转对无缓存片上网络产生的影响.该方法是通过基于对优先级控制算法的改进可以实现拥塞控制,降低在拥塞发生时微片的偏转次数.通过模拟仿真,该方法可以有效降低在高负荷情况微片的偏转次数和平均延迟.

关键词: 无缓存, 片上网络, 低延迟, 低功耗

Abstract:

In low load, bufferless Network-on-Chip (NoC) routing performance is better than buffered routing, and has the advantage of low-power dissipation and low latency. But in the condition of high load due to increased competition and the emergence of deflection decrease bufferless NoC routing causes more time-delay and energy consumption. Based on congestion control, a new way is proposed to reduce packet deflection on bufferless NoC in high load condition. The competition frequency of each output link of the router in the congestion register is recorded. When competition occurs on one output port, a count is increased in the corresponding register. According to the priority policies,while winning the route calculation flits through the effective output port, the other flits are deflected to an inactive output port of low congestion degree. This method, based on improvement of the priority control algorithm, can effectively reduce the number of flits deflection when the congestion occurs. Through the simulation, this method can effectively reduce the deflection and average delay of a flit in high load condition.

Key words: bufferless, NoC, low-latency, low-power dissipation

中图分类号: 

  • TP391

[1] LEE K, LEE S J, YOO H J. Low-power network-on-chip for high-performance SoC design [J]. IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2006, 14(2): 148-160.
[2] ZHANG N, GU H, YANG Y, et al. QBNoC: QoS-aware bufferless NoC architecture [J]. Microelectronics Journal, 2014, 45(6): 751-758.
[3] CRISPÍN G, GÓMEZ M E, PEDRO L, et al. How to reduce packet dropping in a bufferless NoC [J]. Concurrency & Computation Practice & Experience, 2011, 23(1): 86-99.
[4] FALLIN C, CRAIK C, MUTLU O. CHIPPER: A low-complexity bufferless deflectionrouter [C]// IEEE International Symposium on High Performance Computer Architecture. USA, Washington, D C: IEEE Computer Society, 2011: 144-155.
[5] CHRYSOSTOMOU C, TATAS K, RUNCAN A R. A Dynamic fuzzy logic based routing scheme for bufferless NoCs [C]// IEEE International Conference on Computational Science and Engineering. USA, Washington, D C: IEEE Computer Society, 2012: 295-302.
[6] 苑召国, 郭大昌. 超立方体中基于极大安全链路矩阵的容错路由[J]. 广东工业大学学报, 2008, 25(1):33-37.YUAN Zhao-guo, GUO Da-chang. Fault-tolerant Routing Based on Maximum Safety-linkMatrices in Hypercube [J]. JOURNAL OF GUANGDONG UNIVERSITY OF TECHNOLOGY, 2008, 25(1): 33-37.
[7] MOSCIBRODA T, MUTLU O. A case for bufferless routing in on-chip networks [C]//Proceedings of the 36th Annual International Symposium on Computer architecture. New York, NY, USA: ACM Press, 2009: 196-207.
[8] 刘亮亮, 韩国栋, 张帆, 等. 一种无缓存片上网络交叉开关调度机制[J]. 小型微型计算机系统, 2013, 34(7): 1579-1584.LIU L L, HAN G D, ZHANG F, et al. Crossbar scheduling mechanism for bufferless network-on-chip [J]. Journal of Chinese Computer Systems, 2013, 34(7): 1579-1584.
[9] KIM H, KIM Y, KIM J H. Clumsy flow control for high-throughput bufferless on-chip networks [J]. Computer Architecture Letters, 2013, 12(2): 47-50.
[10] CAMACHO J, FLICH J, DUATO J, et al. Towards an efficient NoC topology through multiple injection ports[C]// Proceedings of the 14th Euro micro Conference on Digital System Design. USA, Washington, D C: IEEE Computer Society, 2011: 165-172.
[11] STENSGAARD M B, SPARSO J. Renoc: A network-on-chip architecture with reconfigurable topology[C]// Proceedings of the Second ACM/IEEE International Symposium on Networks-on-Chip. USA, Washington, D C: IEEE Computer Society, 2008: 55-64.
[12] 胥大成, 樊建席, 张书奎. 基于2D-Mesh的容错路由算法[J]. Computer SciXU D C, FAN J X, ZHANG S K. Fault-tolerant routing algorithm in 2D-mesh[J]. Computer Science, 2012, 39(3): 113-117.ce, 2012, 39(3): 113-117.XU D C, FAN J X, ZHANG S K. Fault-tolerant routing algorithm in 2D-mesh[J]. Computer Science, 2012, 39(3): 113-117.
[13] LU Z, ZHONG M, JANTSCH A. Evaluation of on-chip networks using deflection routing[C]// Proceedings of the 16th ACM Great Lakes symposium on VLSI. USA, New York: ACM Press, 2006: 296-301.
[14] LI M, ZENG Q A, JONE W B. DyXY - A proximity congestion-aware deadlock-freedynamic routing method for network on chip[C]// Proceedings of the 43rd annual Design Automation Conference. USA, New York: ACM Press, 2006: 849-852.
[15] XIE Y, YAO L I. NoC system-level modeling based on De Bruijn graph and systemC [J]. Journal of University of Science & Technology of China, 2011, 41(2): 183-188.
[16] YAO C, FENG C, ZHANG M, et al. Low latency multicasting scheme for bufferlesshybrid NoC-bus 3D on-chip networks [J]. Communications in Computer & Information Science, 2015, 491: 36-47.

[1] 孙锋, 刘怡俊. 一种片上网络路由算法的分析与优化设计[J]. 广东工业大学学报, 2017, 34(05): 60-64.
[2] 李峥, 冯永晋, 江志文. 低功耗双模态电子货架标签系统[J]. 广东工业大学学报, 2014, 31(3): 124-129.
[3] 冯永晋, 李峥, 张海笑. 采用相对时间的无线电子货架标签休眠方法[J]. 广东工业大学学报, 2014, 31(3): 130-136.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!