WiredTiger
summary
WiredTiger is an embedded storage engine claiming 7–10x throughput and 80% less storage compared with older engines and powering AWS workloads that report ~50% hardware cost reduction. It uses MVCC with snapshot isolation, a B‑tree core plus optional LSM as a forest of B‑trees, per‑page skip‑list updates, hazard‑pointer pointer swizzling, redo‑only logging with optional no‑journal durability, large pages (128MB design goal), Snappy/LZ4 compression and named checkpoints. Rollback under extreme single‑key contention can reach ~50%.
AI-written summary · about
chapters (33)
- 0:00Intro
- 3:54Deployments
- 5:04MongoDB's Storage Engine API
- 6:41Storage Engine Layer
- 8:27Motivation for Wired Tiger
- 10:00Wired Tiger Architecture
- 11:41Column-store, LSM
- 14:05Traditional B+tree
- 15:07Trees in cache
- 15:58Hazard Pointers
- 19:06Pages in cache
- 20:05Skiplists
- 26:10In-memory performance
- 28:40What is Concurrency Control?
- 29:08Wired Tiger Concurrency Control
- 29:24Multiversion Concurrency Control (MVCC)
- 31:21MVCC In Action
- 32:33Block manager
- 34:26Read path
- 34:42Write path
- 36:03In-memory Compression
- 39:21On-disk Compression
- 41:14Checksums
- 42:34Compression in Action
- 43:12Topics
- 45:40Joumal and Recovery
- 46:59Durability without Journaling
- 48:48Writing a checkpoint
- 49:41Checkpoints in Action (cont.)
- 50:15Named checkpoints
- 52:06What's next for Wired Tiger?
- 55:06Wired Tiger LSM support
- 57:16Benchmarks
related talks
46mUnderstanding and tuning WiredTiger
46mUnderstanding and tuning WiredTiger
28mComparing MongoRocks, WiredTiger and mmapv1 for performance and efficiency
11mAdapting TiKV for Cloud Storage
40mInfluxDB storage engine internals
35mQuestioning Database Claims: Design Patterns of Storage Engine
18mTiFlash: Make TiKV 10x Faster and HTAP-able
13mClosing the B+-tree vs. LSM-tree Write Amplification Gap on Modern Storage Hardware...