FACARS: A novel fully adaptive congestion aware routing scheme for network on chip

摘要

Congestion awareness is deemed a key feature in the design of reliable routing algorithms for network on chip, because of its grave impact on performance. Nonetheless, such task is by no means trivial, as locally aware schemes have a narrow vision on network state, which could lead to poor decision-making, and globally aware routing requires complex computation units and introduces the collection of excessive irrelevant information. In this paper, two versions of a novel reliable fully adaptive congestion aware routing algorithm for network on chip are introduced. The new scheme leverages both local and non-local information in path selection and offers a compromise between locally and globally aware routing. It investigates congestion in parts of the network close, not only to the source, but also to the destination node. Experimental results show that the proposed algorithm outperforms its peers for all configurations in terms of both average latency and throughput.