ASA-routing: A-Star Adaptive Routing Algorithm for Network-on-Chips

摘要

Network congestion is not an uncommon occurrence even when a routing algorithm is well-designed, especially under the condition of a high injection rate. Moreover, it strongly affects the network's overall performance as a result of increased packet latency. However, the majority of existing congestion avoidance methods either utilize local information or are incredibly complicated. The A-star algorithm is characterized as a heuristic algorithm typically used for the purpose of obtaining an optimal path. In this paper, we propose a novel route selection strategy for network-on-chips is proposed. This strategy is based on the A-star algorithm called ASA-routing. This selection method can be coupled with any deadlock-free adaptive routing algorithm. The ASA-routing utilizes routing table information in order to select as non-congested as possible of output channels for forwarding packets. The congestion information should be dynamically updated according to previously routed packets' transmission latency. Based on experimental results for different traffic patterns and network loads, the manner in which our method can be applied to the repetitive turn model routing and the odd-even turn routing is outlined, improving both the average latency and the throughput.