中国科技核心期刊

中文核心期刊

CSCD来源期刊

空间控制技术与应用 ›› 2022, Vol. 48 ›› Issue (1): 58-65.doi: 10.3969/j.issn.1674 1579.2022.01.009

• 短文 • 上一篇    下一篇

面向可重构片上网络的缓存可靠性分析

  

  1. 哈尔滨工程大学计算机科学与技术学院
  • 出版日期:2022-02-26 发布日期:2022-03-07
  • 基金资助:
    黑龙江省自然科学基金项目(JJ2019YX0922)

Cache reliability analysis for reconfigurable network on chip


  • Online:2022-02-26 Published:2022-03-07
  • Supported by:

摘要: 一种新的网络交换体系结构混合电路交换(HCS),已经部署在多处理器系统中,解决了片上网络系统(NoC)所存在的功率过大和较低吞吐量的问题.然而,在HCS中,为保证各个核心数据的一致,采用的是广播形式来解决缓存一致性问题.这样就会占用大量的系统资源,对系统的整体运行会产生较大的影响.因此,试图通过讨论基于HCS的芯片多处理器(CMP)上缓存一致性的实现来解决这个问题,并提出了一种基于k端模型的可靠性协议建模方法.重点研究了三种缓存一致性协议:WriteOnce、MESI和MOESI,并得到了系统可靠性概率的表达式.结果表明,HCS网络与EB网络和总线网络相比较,这些协议的可靠性平均提高了40.22%和59.83%.

关键词: 缓存一致性, 网络交换, 系统可靠性, k-端模型

Abstract: A new network switch architecture, hybrid circuit switched (HCS), is developed for multi processor systems, which solves the problems of over power and low throughput in on chip network systems (NoC). However, in HCS, to ensure the consistency of each core data, the broadcast is adopted to solve the cache consistency problem. This may occupy many system resources, and have a greater impact on system execution. Therefore, in this paper, we try to solve this problem by discussing the implementation of cache consistency on HCS based chip multiprocessor (CMP), and propose a reliability protocol modeling method based on k-terminal model. Three cache consistency protocols, writeonce, MESI and MOESI, are studied, and the expressions of system reliability probability are obtained. The results show that the reliability of these protocols is improved by 40.22% and 59.83% on average in HCS networks compared with elastic buffer (EB) networks and bus networks.

Key words: cache coherence, networking switch, system reliability, k-terminal model

中图分类号: 

  • TP391.4