Bridging the Gap between Relational OLTP and Graph-based OLAP
summary
GART extends HTAP into HTGAP by adding a relational-to-graph mapping layer and an in-memory dynamic graph store that claims tens of milliseconds freshness and multi-order performance gains versus ETL or log‑replicated designs. It uses an efficient mutable CSR split into edge segments, a coarse-grained MVCC epoching for versions, and flexible property placement. On LDBC SNB GART beats LiveGraph by up to 4.4× and reports orders‑of‑magnitude improvement in freshness or throughput.
AI-written summary · about
chapters (11)
- 0:00Bridging OLTP and OLAP
- 2:28Dynamic graph processing solutions
- 4:36Hybrid transactional analytics
- 5:40Gut system design goals
- 7:06Transparent data conversion
- 9:38Dynamic graph storage challenges
- 11:22Storage optimization insights
- 12:26Redesigned graph storage
- 15:56Evaluation methodology
- 17:20Performance comparison
- 19:09Conclusion
related talks
01h 13mHTAP made live — from the engineering perspective
12mTEGRA: Efficient Ad-Hoc Analytics on Evolving Graphs
29mTigerGraph, Connecting the World
29mHTAP vs OLTP/OLAP
41mGraphGen: Conducting Graph Analytics over Relational Databases
29mAll about graphs in Database
30mAnalyze connections and find patterns in data with graphs
36mHow Can HTAP Databases Simplify Your Data Infrastructure?
