Explore a conference talk that delves into Seraph, a groundbreaking fully-external graph computation system designed to achieve optimal scalability and improved efficiency. Learn how Seraph addresses the challenges of processing large-scale graphs with limited memory on a single machine by implementing on-demand processing techniques. Discover the innovative approaches used to balance I/O operations and access locality, including a hybrid format for graph data representation, efficient vertex update mechanisms, and selective pre-computation strategies. Gain insights into how Seraph outperforms other state-of-the-art fully-external systems when handling billion-scale graphs and various graph algorithms, demonstrating performance improvements of up to two orders of magnitude.
Overview
Syllabus
FAST '24 - Seraph: Towards Scalable and Efficient Fully-external Graph Computation via On-demand...
Taught by
USENIX