G-CARA: A Global Congestion-Aware Routing Algorithm for traffic management in 3D networks-on-chip

摘要

Congestion-aware routing algorithms improve the network performance by selecting less congested paths between source and destination routers. In this paper, a Global Congestion-Aware Routing Algorithm (G-CARA) is proposed for traffic management in three-dimensional Networks-on-Chip (3D NoCs). The proposed method forwards data packets efficiently toward regions with lower traffic to avoid network congestion. This is done by two steps. At first, the algorithm collects global congestion information in each layer of the network, and then, the algorithm utilizes the incoming information to specify an optimal path for sending a packet based on dimension-order routing. Global congestion values are propagated by embedded traffic information in header flit of packets. Furthermore, turn models are used to guarantee deadlock-freedom in proposed routing algorithm. Experimental results show that G-CARA can achieve an improvement in both saturation point and average latency within the range of 14%-22% and 42%-61% on average, respectively.