AI summaryⓘ
The authors developed SIGMA, a new method to split large graphs for training Graph Neural Networks (GNNs) across multiple computers. Unlike previous methods that focus on either dividing vertices or edges and often optimize only one goal, SIGMA can handle both with multiple objectives and constraints at once. It also uses a clustering step to understand the graph's overall structure, improving how it splits the graph while staying efficient. When tested on various graphs and GNN systems, SIGMA showed strong performance, balancing communication, computation, and memory needs better than many existing methods. This suggests that a single flexible method can work well for different distributed GNN training setups.
Graph Neural NetworksGraph PartitioningVertex PartitioningEdge PartitioningDistributed TrainingStreaming PartitioningEdge CutVertex CutPartition BalanceClustering
Authors
Barbara Hoffmann, Shai Dorian Peretz, Adil Chhabra, Ahmet Kadir Yalcinkaya, Ruben Mayer, Christian Schulz
Abstract
Distributed Graph Neural Network (GNN) training depends critically on how the underlying graph is partitioned across compute resources. Existing graph partitioners focus either on vertex partitioning or edge partitioning and typically optimize only a single communication objective (edge cut or vertex cut) under a single balance constraint (vertex balance or edge balance). We present SIGMA (Streaming Integrated Graph Partitioning with Multi-objective Awareness), a versatile streaming graph partitioner that supports both vertex and edge partitioning within a unified multi-objective, multi-constraint framework. Depending on the target distributed GNN system, SIGMA can be configured for edgecut-oriented vertex partitioning or vertex-cut-oriented edge partitioning while simultaneously accounting for both vertex and edge balancing. A clustering-based preprocessing stage incorporates global graph structure to improve partition quality while preserving the efficiency and scalability advantages of streaming partitioning. We evaluate SIGMA on six benchmark graphs spanning diverse domains and scales using two distributed GNN training systems: Dist-GNN (edge-partitioned) and DistDGL (vertex-partitioned). Across both settings, SIGMA consistently achieves strong performance, showing its ability to navigate complex trade-offs between partition quality, training efficiency, and memory consumption, frequently outperforming streaming baselines while remaining competitive with high-quality in-memory partitioners such as METIS, KaHIP and HEP. These results demonstrate that a unified streaming partitioner can effectively address the communication, compute, and memory challenges of distributed GNN training across fundamentally different system architectures.