Leveraging Moore's Law to Optimize Database Performance
summary
Evaluates Moore’s Law against present hardware and explains how Aerospike uses much more RAM per disk plus NVMe to change design tradeoffs, keeping a full primary index in RAM instead of a cache and using balanced-tree lookups to map straight to disk. Demonstrates a four-node laptop cluster achieving 400,000 reads/s 200,000 writes/s P99 ≈ 1ms P999 ≈ 1ms on random workloads, claims predictable submillisecond tails, and contrasts that with cache hit/miss models and their limited end-to-end benefits.
AI-written summary · about
chapters (11)
- 0:00Moore's Law and database performance
- 1:18Moore's Law and transistor growth
- 2:43CPU scaling limits and scalability
- 4:28Databases and primary indexes
- 8:02Hardware evolution 2006–2024
- 13:28RAM versus disk capacity trends
- 15:33Traditional caching limitations
- 18:45Cache hit rates and app performance
- 20:55Full index in RAM for uniform latency
- 23:24Aerospike in-memory indexing results
- 25:09Summary
related talks
26mLeveraging Moore's Law to Optimize Database Performance
22mThe Art of the Possible
11mThe Power of Multi-Model: KV, Document, Graph
28mAerospike's Journey Through Innovations, Sustainability, and Roadmap
45mAerospike + Persistent Memory = Voila!
15mAWS On Air Voice of the Customer ft. Aerospike, Inc.
32mFuture-Proofing your Architecture - AI at Any Scale with HPE & Aerospike
24mCloud Workloads At Petabyte Scale