Leveraging Moore's Law to Optimize Database Performance
summary
Argues that modern CPU core counts, large RAM and NVMe change tradeoffs so the primary index lives fully resident in RAM rather than as a cache, using a balanced-memory index to pinpoint on-disk pages. Demonstration with Aerospike NoSQL on a four-node laptop cluster achieves about 400k reads/sec and 200k writes/sec with p99 and p999 ≈ 1 ms and worst reads under 2 ms and worst writes under 4 ms, showing the design actually works.
AI-written summary · about
chapters (8)
- 0:00Moore's Law and database performance
- 3:27CPU speed to scalability shift
- 4:53Primary indexing and data structures
- 8:02Hardware advancements 2006-2024
- 14:32Traditional caching limitations
- 20:55In-memory primary indexes optimization
- 23:04Aerospike uniform response times
- 24:48Summary
related talks
26mLeveraging Moore's Law to Optimize Database Performance
11mThe Power of Multi-Model: KV, Document, Graph
28mAerospike's Journey Through Innovations, Sustainability, and Roadmap
15mAWS On Air Voice of the Customer ft. Aerospike, Inc.
24mCloud Workloads At Petabyte Scale
22mThe Art of the Possible
46mEdge Performance with In memory NoSQL
16mAerospike + Intel® persistent memory