Exploring Graph Partitioning Techniques for GNN Processing on Large Graphs: A Survey

Graph Neural Networks (GNNs) have evolved as a powerful tool for understanding and processing graphical data. However, their effectiveness is often hindered by the computational challenges posed by large-scale graphs. In order to mitigate these challenges, graph partitioning techniques have been widely employed to cut a large graph into smaller, manageable subgraphs. This survey paper provides a comprehensive analysis of graph partitioning methods specifically tailored for GNN processing on large graphs. We explore a wide range of partitioning algorithms and strategies, including clustering, multi-level graph partitioning, and community detection approaches. Furthermore, we investigate the impact of different partitioning criteria, such as load balancing, communication overhead, and preservation of graph properties, highlighting the importance of preserving connectivity, neighborhood information, and graph semantics. Through an extensive review of the literature, the strengths and limitations of existing graph partitioning techniques are identified and propose potential avenues for future research.