This Systematization of Knowledge paper was completed in a group of two after reading 20 relevant research and review papers for the class ECE 226.
Abstract: Distributed Graph Neural Network (GNN) acceleration strategies are crucial for learning large GNNs on complex graph datasets. In this systemization of knowledge paper, we contribute to the field by providing a comprehensive and analytical comparison of recent advancements. We classify information by acceleration mechanism, compare and rank recent systems and their performances, and draw valuable conclusions to guide future research and explore unexplored areas. Overall, we find that the current best acceleration systems are BGL, P3, and MGG.