RonDB: A Key-Value Store with SQL Capabilities and LATS Properties
summary
A distributed key-value engine with SQL frontends claims 100–200 μs lookups at high load and targets under 30 seconds downtime per year by using up to 64 data nodes, per-CPU hundreds of thousands of lookups and cluster-level billions/sec. It exposes MySQL, C++, Java, LDAP (and soon REST) APIs, supports 16 TB in-memory with hundreds of TB of disk, uses adaptive checkpointing, adaptive CPU spinning and row-based partial checkpoints that write ~80× less than page checkpoints, plus a non-blocking two‑phase commit via coordinator rebuild.
AI-written summary · about
chapters (38)
- 0:00Introduction
- 4:00Requirements
- 5:54Latency Requirements
- 8:35RonDB Architecture
- 9:38Use Cases
- 10:17DNS Server
- 10:38Stock Exchange
- 12:50Uptime
- 15:13Availability Zones
- 17:18Change Cluster Size
- 18:29Change Cluster Interface
- 19:13Node Groups
- 24:33Data structure
- 25:38Indexes
- 26:34Ttrees
- 28:51Virtual Machine
- 30:21Memory Architecture
- 31:29Thread Architecture
- 33:25Query Threads
- 37:48B Resources
- 38:38Partial Checkpoints
- 39:32Full Checkpoints
- 39:52InMemory Checkpoints
- 40:46Memory Sizes
- 41:51Test Case
- 42:44PageBased Checkpoint
- 43:42RowBased Checkpoint
- 45:04Checkpoint Speed
- 46:26Checkpoint Complexity
- 47:28Global Checkpoint
- 47:58Checkpoint Consistency
- 51:03Checkpoint Optimization
- 52:25Questions
- 53:13Adaptive CPU Spinning
- 55:18Networking
- 55:30Memory buffers
- 56:23Memory barrier protocol
- 57:42Wrapup
related talks
12mSpanDB: A Fast, Cost-Effective LSM-tree Based KV Store on Hybrid Storage
11mEvolution of Development Priorities in Key-value Stores Serving Large-scale Applications
29mDistributed SQL: How TiDB Modernized Data Management
29mThe Design and Evolution of Event-Driven Databases
29mDistributed SQL How TiDB Modernized Data Management
16mScaleDB: A Scalable, Asynchronous In-Memory Database
14mYour favourite second database
55mThe Future of MySQL