סמינר: Graduate Seminar
Acceleration of Large-Scale GNN Training using SmartNICs
Date:
October,01,2024
Start Time:
14:00 - 15:00
Location:
1061, Meyer Building
Zoom:
Zoom link
Add to:
Lecturer:
Liad Gerstman
Research Areas:
| Graph Neural Networks (GNNs) have emerged as powerful algorithms for tackling complex problems across diverse domains, including biology, social networks, chip design, and recommendation systems. However, the increasing size of graph datasets poses significant challenges for traditional single-node CPU-GPU systems, making it difficult to store and process these graphs efficiently. In this lecture, I will delve into the challenges of large-scale GNNs and introduce LGNNIC—an innovative SmartNIC-based inter-node system architecture accelerating large-scale GNN training. By leveraging the in-network programmability of SmartNICs, LGNNIC enables efficient graph sampling near remote memory nodes, reducing the data transfer overhead to local computational nodes. I will present a detailed timing analysis to demonstrate the acceleration benefits of this approach, examine key trade-offs, and outline future research directions.
M.Sc. student under the supervision of Prof. Avi Mendelson.
|