GraphHouse and Training GNNs at Internet Scale
Title: Introducing GraphHouse: A Cloud Native Graph Database Designed For Datalake
In this talk, we unveil TigerGraph's new cloud-native product powered by the cutting-edge GraphHouse architecture, a fusion of Graph analytics and data lakes. GraphHouse enables seamless data ingress from Delta Lake, Iceberg, Snowflake, Postgres, BigQuery, etc., utilizing vertices and edges to store data directly on cost-effective, reliable cloud object storage. By decoupling the compute layer from the data layer, users can dynamically spin up compute clusters on-demand, accessing a single connected dataset. All graph database capabilities, including data silo integration, fast multi-hop traversals, and query visualizations, are seamlessly inherited, with unlimited compute capacity.
Speaker : Mingxi Wu (TigerGraph)
Mingxi Wu is the Head of Engineering at TigerGraph. His career and passions have been dedicated to data management, with work experience at Microsoft’s SQL Server Group, Oracle’s Relational Database Optimizer Group, and Turn Inc.'s Big Data Management Group. He has received five prestigious research awards from SIGMOD, KDD, and VLDB premier conferences, and has authored numerous patents in graph database systems. Mingxi received his PhD from the University of Florida, specialized in both database and data mining.
Talk 2: Training GNNs at Internet Scale using cuGraph and WholeGraph
We present our approach to manage 70TB graph datasets, and train GraphSage across 1024 GPUs. One key feature of our approach is the separation of the graph sampling and GNN training phases, giving the user flexibility to scale each independent of the other. WholeGraph provides a distributed feature store that leverages GPU memory and caching to provide high performance dataloading. Dataloading and sampling are the two largest bottlenecks in GNN training according to our profiling.
Speaker: Joe Eaton (NVIDIA)
Joe Eaton is a Distinguished System Engineer for Data and Graph Analytics at NVIDIA, and is currently leading the company strategy for Graph Neural Networks at Nvidia. Joe leads teams for code optimization, graph analysis, framework development and optimization,
as well as interacts with prospective customers in industry.
Key areas of interest are financial services, retail Recommenders, and molecular generation for drug discovery.