ScaleDB: A Scalable, Asynchronous In-Memory Database
summary
A working in‑memory DB is built that delays range index updates and applies them in batches using per‑thread flushes at merge epochs, storing committed writes in small hash indexlets instead of immediately mutating trees. A new optimistic protocol called ACC validates reads while range indexes remain stale and detects phantoms with compact phantomlets. On a dual‑socket 36 core server it implements the idea and shows low abort rates and 1.8× TPC‑C throughput vs Cicada and 9.5× YCSB vs Peloton.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 0:28Scalability challenges
- 1:30Contention in range indexes
- 2:33Contention in transactions
- 3:39Separating guarantees
- 4:42Scan latency analysis
- 6:26Design principles
- 6:49Indexlets for updates
- 7:52Asynchronous updates in action
- 8:55Asynchronous Concurrency Control
- 9:39Handling phantoms
- 11:43Evaluation vs Cicada
- 12:46TPC-C scalability
- 14:54Conclusion
related talks
46mEdge Performance with In memory NoSQL
31mReimagining the Future: Where the Market is Evolving
29mThe Design and Evolution of Event-Driven Databases
29mPurpose-built for performance: Amazon MemoryDB for Redis
01h 02mRonDB: A Key-Value Store with SQL Capabilities and LATS Properties
42mThe hows and whys of a distributed SQL database
12mSpanDB: A Fast, Cost-Effective LSM-tree Based KV Store on Hybrid Storage
41mPerformance & Fault Tolerance: Building a Modern Database