Overview
Explore a conference talk from FAST '19 on optimizing large-scale graph processing for emerging storage devices. Learn about the challenges of graph processing on huge datasets and how traditional solutions become inefficient with modern storage technologies like SSDs. Discover a new graph partitioning and processing framework designed to leverage the capabilities of these devices, offering up to 2X performance improvement over state-of-the-art solutions. Gain insights into fine-grained access in external graph processing, partitioning strategies for vertex data, and experimental results demonstrating the effectiveness of this approach. Understand the potential impact on applications analyzing massive datasets and the future of graph processing architectures.
Syllabus
Intro
Large-Scale Graph Processing Challenges
Fine-Grained Access in External Graph Processing
Programming Model
Prior External Graph Processing -- Graf Boost
Scalability Issue
Partitioning Graph Data
Instead, We Propose a Partitioning for Vertex Data
Execution Flow
Updating Vertex Mirrors on Different Partitions
Experimental Setup
Performance Evaluation
Execution Time Breakdown
Concluding Remarks
Taught by
USENIX