Graphs as Streams: Rethinking Graph Processing in ... #bbuzz
summary
Argues that graphs do not need full storage to be analyzed and demonstrates a working prototype built on Apache Flink called jelly stream that runs single-pass, distributed algorithms using compact summaries. Shows an exact streaming connected components implementation using a union‑find style summary with periodic merges and time windows, and provides approximate methods (spanners, sparsifiers, neighborhood sketches) for bipartiteness, triangle estimates and distance estimation. Deletions are harder but additions stream efficiently.
AI-written summary · about
chapters (10)
- 0:00Graph streaming introduction
- 1:33Traditional graph processing
- 3:35Graph streaming challenges
- 5:35Graph summaries and models
- 6:50Connected components algorithms
- 8:40Streaming implementation
- 10:18Modern streaming challenges
- 11:42Gelly-stream prototype
- 13:48Gelly-stream features
- 14:29Conclusion and Q&A
related talks
01h 21mAn Overview on Data Streams Algorithms
32mBig Data Processing : Where do we go next ?
28mSweet Streams Are Made Of These
42mPrinciples and Patterns for Streaming Data Analysis
49mscale.bythebay.io: Stephan Ewen, Apache Flink and the Next Wave of Stream Processing Applications
42mApache Flink deep-dive #bbuzz
43mTo Petascale and Beyond: Apache Flink in the Clouds
43mThe Stream Processor as a Database