MiniCAR: Minimal Congestion-Aware Routing Method in Fine-Grained Circuit-Switched Networks for Parallel Computing Systems

In parallel high-performance computing (HPC) systems, network congestion is one of the main factors to degrade communication performance, because it may lead to increased end-to-end latency and power consumption. In this work, we address this issue by the means of a simple routing algorithm on a target fine-grained circuit-switched (FGCS) network. The number of allocated slots for each FGCS switch in the network is a direct factor to affect the end-to-end latency. Our proposed approach employs a minimal congestion-aware routing (MiniCAR) method to perform better routing decisions and alleviate the network congestion so that the minimum necessary number of slots can be reduced in a target FGCS network. Evaluation results show that, compared to the traditional dimension order routing algorithm, MiniCAR occupies a smaller number of time slots by up to 50.8% on a 2-D torus interconnection network.