Reproduce “Ripple: Scalable Incremental GNN Inferencing on Large Streaming Graphs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pranjal Naman, Yogesh Simmhan
Ripple provides a scalable and exact framework for incremental Graph Neural Network (GNN) inference on large streaming graphs. It efficiently propagates changes for GNNs employing linear aggregation functions, achieving up to 2000x faster throughput compared to DGL's CPU-based layer-wise recompute and 30x higher throughput in distributed settings for web-scale graphs.
Key insights
- Propagating only incremental "deltas" rather than full recomputations significantly reduces computational overhead for GNN inference on dynamic graphs.
- The linearity of common GNN aggregation functions (sum, mean, weighted sum) enables exact incremental updates by simply summing delta messages accumulated in per-vertex mailboxes.
- Combining this delta propagation with a distributed architecture and efficient communication strategies allows GNN inference to scale to graphs exceeding single-machine memory, maintaining accuracy and performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.