Seraph: Towards Scalable and Efficient Fully-external Graph Computation via On-demand...
summary
Seraph is a fully-external graph engine that replaces streaming-within-grid I/O with on-demand processing tuned for modern SSDs and claims up to two orders of magnitude speedup on billion-scale graphs (testing includes a 1.7 TB graph). It implements a hybrid format that stores dense blocks as grids and sparse blocks as rows, a vertex passing log-and-merge scheme to avoid random writes and preserve locality, and selective pre-computation to reuse loaded blocks and cut total I/O — experiments show large runtime and I/O reductions.
AI-written summary · about
chapters (11)
- 0:00Seraph introduction
- 1:21Graph systems background
- 4:55Motivation for Seraph
- 7:47On-demand vs streaming
- 11:18Seraph design principles
- 14:08Hybrid graph storage
- 15:10Vertex passing optimization
- 16:35Selective computation
- 18:21Evaluation setup
- 19:44Memory scalability results
- 20:49Conclusion
related talks
32mLarge-Scale Graph Processing on Emerging Storage Devices
27mData @scale Seattle - Yucheng Low
20mGraphWalker: An I/O-Efficient and Resource-Friendly Graph Analytic System...
09mEfficient Large Graph Processing with Chunk-Based Graph Representation Model
39mThorny path to the Large Scale Graph Processing
24mLarge Scale Graph Computation with Apache Giraph
50mBig Data on Smallish Machines
24mGLogS: Interactive Graph Pattern Matching Query At Large Scale