A framework for designing congestion-aware deterministic routing

  • Authors:
  • Abbas Eslami Kiasari;Axel Jantsch;Zhonghai Lu

  • Affiliations:
  • Royal Institute of Technology (KTH), Sweden;Royal Institute of Technology (KTH), Sweden;Royal Institute of Technology (KTH), Sweden

  • Venue:
  • Proceedings of the Third International Workshop on Network on Chip Architectures
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper, we present a system-level Congestion-Aware Routing (CAR) framework for designing minimal deterministic routing algorithms. CAR exploits the peculiarities of the application workload to spread the load evenly across the network. To this end, we first formulate an optimization problem of minimizing the level of congestion in the network and then use the simulated annealing heuristic to solve this problem. The proposed framework assures deadlock-free routing, even in the networks without virtual channels. Experiments with both synthetic and realistic workloads show the effectiveness of the CAR framework. Results show that maximum sustainable throughput of the network is improved by up to 205% for different applications and architectures.