Back to Search Start Over

Adaptive distribution of control messages for improving bandwidth utilization in multiple NoC.

Authors :
Yadav, Sonal
Laxmi, Vijay
Kapoor, Hemangee
Gaur, Manoj Singh
Kumar, Amit
Source :
Journal of Supercomputing; Oct2023, Vol. 79 Issue 15, p17208-17246, 39p
Publication Year :
2023

Abstract

The advancement of networks-on-chip (NoCs) is noteworthy as the number of cores increases. The bandwidth demand has grown steadily as network traffic has increased owing to high-workload applications. The NoC traffic broadly divided into control messages and data messages in which data messages are bigger in size. As NoC channel bandwidth sets in proportion to the size of the data messages, the NoC bandwidth remains underutilized during control messages transmission. This adversely affects NoC power and performance efficiency. In modern NoC architectures, multiple NoC is popular to efficiently utilize NoC bandwidth because it offers more than one physical channel for traffic communication. The conventional multiple-NoC architectures statically distribute traffic between the NoCs. This significantly affects the power-performance metrics. We have observed up to fivefold variation in energy efficiency during the analysis of static traffic distribution for multiple NoC. In this paper, we propose an adaptive distribution of control messages for multiple NoC to improve bandwidth utilization. The traversal of control messages switch between the NoC networks according to the runtime utilization of networks. The proposed adaptive distribution of control messages improves energy efficiency up to 72.7 % and 66.9 % on average over single-NoC and static traffic distribution in multiple NoC, respectively. The link utilization also improves by 1.37 × and 40 % on average over single-NoC and conventional static traffic distribution, respectively. Thus, the proposed adaptive distribution overcomes the implications of static traffic distribution. [ABSTRACT FROM AUTHOR]

Details

Language :
English
ISSN :
09208542
Volume :
79
Issue :
15
Database :
Complementary Index
Journal :
Journal of Supercomputing
Publication Type :
Academic Journal
Accession number :
171101281
Full Text :
https://doi.org/10.1007/s11227-023-05208-0