Explore a groundbreaking conference talk on scaling graph databases to unprecedented levels. Delve into the Graph Database Interface (GDI), a novel approach that leverages HPC practices to achieve exceptional performance and scalability. Learn about the challenges in graph database development and how GDI addresses them through a portable and programmable API specification. Discover the implementation of GDI for distributed-memory RDMA architectures, utilizing one-sided RDMA communication and collective operations. Examine the impressive results, including scaling to over 100,000 cores for both OLTP and OLAP workloads. Gain insights into the future of extreme-scale graph databases and their potential impact on academic and industry applications.
The Graph Database Interface: Scaling Online Transactional and Analytical Graph Workloads
Scalable Parallel Computing Lab, SPCL @ ETH Zurich via YouTube
Overview
Syllabus
Introduction: Labeled Property Graph and Graph Database Workloads
State of Challenges & Problems
Our Objectives
Graph Database Interface
GDI Standard
GDI-RMA - GDI Implementation for RDMA Systems
GDI-RMA Data Layout
GDI-RMA Example Graph Update Protocol
Evaluation: Low Latency
Evaluation: Scaling OLTP to 120,000 Cores
Evaluation: Scaling OLAP to 120,000 Cores
Conclusion
Taught by
Scalable Parallel Computing Lab, SPCL @ ETH Zurich
