Graph Systems

Large-Scale GNN Training

Gradient-only communication for large-scale GNN training

Active Research

Overview

Grappa trains graph neural networks by exchanging only gradients between partitions and periodically repartitions the graph to expose new neighborhoods. In the paper's experiments it trains 4x faster on average (up to 13x) than the baseline systems, and it scales to trillion-edge graphs on commodity hardware.

Related Publications

Grappa (arXiv 2026)
GNN Partitioning Survey (GRADES-NDA 2025)