Scalable load balancing congestion-aware Network-on-Chip router architecture

  • Authors:
  • Chifeng Wang;Wen-Hsiang Hu;Nader Bagherzadeh

  • Affiliations:
  • Dept. of Electrical Engineering and Computer Science, University of California, Irvine, Irvine, CA 92697-2625, USA;Dept. of Electrical Engineering and Computer Science, University of California, Irvine, Irvine, CA 92697-2625, USA;Dept. of Electrical Engineering and Computer Science, University of California, Irvine, Irvine, CA 92697-2625, USA

  • Venue:
  • Journal of Computer and System Sciences
  • Year:
  • 2013

Quantified Score

Hi-index 0.00

Visualization

Abstract

Adaptive routing algorithms have been employed in interconnection networks to improve network throughput and provide better fault tolerance characteristics. However, they can harm performance by disturbing any inherent global load balance through greedy local decisions. This paper proposes a novel scalable load balancing congestion-aware Network-on-Chip (NoC) architecture that not only enhances network transmission performance while maintaining a feasible implementation cost, but also improves overall network throughput for various traffic scenarios. This congestion control scheme which consists of dynamic input arbitration and adaptive routing path selection is proposed to balance global traffic load distribution so as to alleviate congestion caused by heavy network activities. Furthermore, faulty links information can be broadcasted by existing congestion management control signals to prevent packets from routing through defected areas in order to eliminate potential heavy congestion situations around these regions. Experimental results show that throughput is improved dramatically while maintaining superior latency performance for various traffic patterns. Compared to a baseline router, the proposed congestion management mechanism requires negligible cost overhead but provides better throughput for both mesh and diagonally-linked mesh NoC platforms.