Redwood Storage Engine Update
Talkat FoundationDB Summit 2019 · on Monday, 2019-11-18· at 11:40· Grand Ballroom 8, Marriott Marquis San Diego Marina·
summary
Redwood is a B+ tree storage engine for FoundationDB that ships in recent 6.2 patches and targets higher I/O throughput and smaller on-disk footprints. It implements multi-version pages, prefix compression across and within nodes, a pager with a new delayed write-ahead log to bound page-table memory, and deferred range clears that show about 100x less IO in tests (40M records, 100k prefixes, 32B values; load ~116s, ~1.5GB footprint with 2GB cache). Remaining work: CPU optimizations, node merging, stable on-disk format; first production mode is FDB 7.0 single-version only.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:35Redwood Overview
- 1:12Status - What's done?
- 1:47Status - What's not done?
- 2:37Prefix Compression Performance
- 6:42Pager - Original Design
- 8:38Pager - Updated Design
- 11:16I/O Patterns - Writes Redwood
- 14:57VO Patterns - Large GetRange
- 16:51I/O Patterns - Deferred Range Clear
- 19:05Future Improvements
- 20:25Conclusion
related talks
26mFuture of FoundationDB Storage Engines
22mSolving Everyday Data Problems with FoundationDB
11mEvolution of Development Priorities in Key-value Stores Serving Large-scale Applications
26mRunning FDB at scale at Wavefront
57mRocksDB
34mFoundationDB, the Black Knight
13mClosing the B+-tree vs. LSM-tree Write Amplification Gap on Modern Storage Hardware...
51mSix New Important RocksDB Features And Planned Works