Re-think Data Management Software Design Upon the Arrival of Storage Hardware with
Talkat 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage '20) · 2020-07-13..07-14·
summary
Argues that transparent compression embedded in drives (an FPGA-based flash controller exposing a standard block device) decouples logical from physical capacity so software can waste logical space for simpler, faster designs. Shows Postgres can drop fillfactor aggressively (example: 100→15), gaining ~33% TPS on Sysbench TPCC while physical bytes on the compressed drive barely increase (examples: logical 1.4 TB maps to ~342→365 GB physical). Implements a table-less hash KV store prototype that beats RocksDB with >2x throughput and >2x lower CPU.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:24The Rise of Computational Storage
- 1:31Computational Storage: A Simple Idea
- 2:14ScaleFlux Computational Storage Drive
- 3:06CSD 2000: Data Path Compression/Decompression
- 4:19Open a Door for System-level Innovations Data path compression
- 7:45PostgreSQL (Sysbench-TPCC)
- 9:14Table-less Hash-based Key-Value Store
- 12:55Summary
related talks
28mHow Can Databases Capitalize on Computational Storage?
13mClosing the B+-tree vs. LSM-tree Write Amplification Gap on Modern Storage Hardware...
34mIn core compression-How to shrink your database
37mKEYNOTE: Data Architecture Turned Upside Down
35mQuestioning Database Claims: Design Patterns of Storage Engine
42mCan File Systems Survive in Data-centric World?
16mPolarStore: High-Performance Data Compression for Large-Scale Cloud-Native Databases
27mPreview: Storage