An efficient fault-tolerant routing algorithm in NoCs to tolerate permanent faults
详细信息 查看全文
- 作者：Reza Akbar ; Ali Asghar Etedalpour ; Farshad Safaei
- 关键词：Fault ; tolerant ; Networks ; on ; chip ; Mesh topology ; Routing algorithm ; Permanent faults ; Fault region ; Deadlock ; Power consumption
- 刊名：The Journal of Supercomputing
- 出版年：2016
- 出版时间：December 2016
- 年：2016
- 卷：72
- 期：12
- 页码：4629-4650
- 全文大小：1,576 KB
- 刊物类别：Computer Science
- 刊物主题：Programming Languages, Compilers and Interpreters
Processor Architectures
Computer Science, general - 出版者：Springer Netherlands
- ISSN：1573-0484
- 卷排序：72
文摘
With the possibility of integrating multiple cores into a single chip, research on the networks-on-chip (NoCs) as a kind of interconnection network has assumed great significance. In such networks, the effort is to provide broadband and extendable infrastructure for multi-core architectures. Communication between processors in an NoC is established using routing algorithms. Meanwhile, NoCs, like any other system, are prone to failure. With the increase in the number of network components on a chip, the probability of failure increases, too. Therefore, considering a fault-tolerant mechanism in NoCs seems to be a necessity. The main challenge of this work is combining performance and fault tolerance while reducing power, complexity and cost. In this paper, a fault-tolerant routing algorithm for tolerating static and dynamic faults in 2D Mesh NoCs and node failure model is presented. It should be taken into consideration that despite many other routing algorithms, the proposed method uses only one Virtual Channel. Results show that this method has lower latency and power consumption than SAVA and segment-based (SB) routing algorithms. It showed 2.91 and 12.74 % less power consumption than SAVA and SB, respectively, under SPLASH-2 traffic in an 8 \(\times \) 8 Mesh network with 8 faulty nodes. Its average latency, under Uniform, Transpose, and SPLASH-2 traffics in a 4 \(\times \) 4 Mesh with 4 faulty nodes and an 8 \(\times \) 8 Mesh network with 8 faulty nodes, was reduced by 4.39 and 14.08 % compared to SAVA and SB, respectively.