A Low-latency NoC Router Priority Scheme for BFS algorithm

Breadth-first search (BFS) is one of the most common used graph traversal algorithms and a building block for various graph applications. Many studies have been done to accelerate BFS-based graphic algorithm on multi-core architecture that based on Network-on-Chip NoC. However, BFS is known to be an algorithm that is difficult to obtain better performance from parallelization. One of the most important indicators to measure performance of BFS on a multi-core architecture is the latency of the NoC. In order to reduce the on-chip network delay, this paper proposes a priority strategy based on the global hop count of the packet. Experimental results by simulation show that this strategy can reduce the average packet delay by up to 21%.