GNNAdvisor: An Adaptive and Efficient Runtime System for GNN Acceleration on GPUs¶

Wang, Y., Feng, B., Li, G., Li, S., Deng, L., Xie, Y., & Ding, Y. (2021). {GNNAdvisor}: An Adaptive and Efficient Runtime System for {GNN} Acceleration on {GPUs}. 515–531. https://www.usenix.org/conference/osdi21/presentation/wang-yuke

真的好长啊...

1. First, GNNAdvisor explores and identifies several performance-relevant features from both the GNN model and the input graph, and uses them as a new driving force for GNN acceleration.
2. Second, GNNAdvisor implements a novel and highly-efficient 2D workload management, tailored for GNN computation to improve GPU utilization and performance under different application settings.
3. Third, GNNAdvisor capitalizes on the GPU memory hierarchy for acceleration by gracefully coordinating the execution of GNNs according to the characteristics of the GPU memory structure and GNN workloads.
4. Furthermore, to enable automatic runtime optimization, GNNAdvisor incorporates a lightweight analytical model for an effective design parameter search.