How Can Databases Capitalize on Computational Storage?
summary
ScaleFlux CSD 2000 drives run compression and other compute in-drive so MySQL and Postgres see about 2x throughput versus NVMe with host compression while holding storage footprint down 50–80 percent. Lowering Postgres fillfactor from 100 to 75 raises TPC‑C TPS ~33 percent with physical size only edging from 340 to 360 GB on the CSD. A sparse logging layout cuts write amplification by over 90 percent for tiny transactions and a prototype Cadillac DB hits ~2.5× throughput with 2–3× less CPU. transparent compression
AI-written summary · about
chapters (13)
- 0:00Intro
- 1:49What is a Computational Storage Drive CSD
- 2:05Why use Computational Storage?
- 3:43Benefits of Computational Storage with Data
- 4:32Realizing the Benefits of CSDs for Databases.
- 5:51Transparent Compression: Performance
- 7:07Transparent compression: Cost Industry Research
- 8:14Transparent compression: Combining Cost Savings with Performance Enhancements
- 10:10Open the Door for System Optimization and Innova
- 11:47Optimization: Cost & Performance
- 15:55Minor Tweaks: Redo/Binary Log Current Problem
- 20:50Minor Tweaks: Dual-Mode Page Format
- 22:36Innovation: Sparse Data Structures Table-less Hash-based KV Store
related talks
14mRe-think Data Management Software Design Upon the Arrival of Storage Hardware with
34mIn core compression-How to shrink your database
20mMaximizing the Benefit of Flash Storage for Aerospike
19mDo You Believe in Magic?
PolarStore: High-Performance Data Compression for Large-Scale Cloud-Native Databases
06mScaling your applications with Cloud SQL for MySQL
20mExtreme Performance and SmartScaling MySQL with Storage Class Memory Servers
41mUsing Databases to Pull Your Application Weight