What We Talk About When We Talk About Disk IO
summary
Compares concrete trade-offs between LSM trees and B‑trees, explains why Cassandra and RocksDB use SSTables and how compaction works, measures which patterns cause random vs sequential IO on HDDs and SSDs, and states the ROM conjecture linking read, write and memory overheads. Covers costs: buffered sequential writes, multi-file reads, tombstones, and compaction stalls; lists practical mitigations (batched flushes, bloom filters, compression) so viewers can pick a storage design for write‑heavy vs read‑heavy workloads.
AI-written summary · about
chapters (6)
- 0:00Introduction and motivation
- 2:16Key IO access patterns
- 6:51Log-structured merge trees
- 15:18B-trees overview
- 27:49Trade-offs in storage design
- 32:30Evaluating storage systems
related talks
01h 16mDeep Dive into On-Disk Data Structures
45mThe Database: A Sophisticated and Special Data Structure
52mHow shit works: Databases
48mDesign of data-intensive applications
13mClosing the B+-tree vs. LSM-tree Write Amplification Gap on Modern Storage Hardware...
18mPreview: KV-Stores and Databases
35mQuestioning Database Claims: Design Patterns of Storage Engine
01h 16mOn-Disk Data Structures: B-Trees, LSM-Trees and Fractal Trees