MyFlashSQL: Flash is more than faster-harddisk
summary
MyFlashSQL modifies the MySQL engine and shows concrete gains: changing buffer page size from 16KB to 4KB yields ~2.5x throughput on TPC-C/LinkBench; reworking secondary-index scans to issue submit_io parallel reads exploits SSD chip parallelism and shows 3.5–10x single-thread speedups on consumer Intel/Samsung drives; adding a SHARE SSD write interface eliminates double-write and cuts write volume ~50% with ~2–2.4x speedup; switching to a late-write low paradigm improves read latencies up to ~3.7x and gives ~30% on high-end SSDs.
AI-written summary · about
chapters (10)
- 0:00MyFlashSQL introduction
- 1:10Flash storage characteristics
- 3:24SSD impact on databases
- 5:33Page size for SSD
- 7:25Index scan with SSD parallelism
- 9:58Double buffer optimization
- 14:22Write-after-read to read-after-write
- 18:01Low mode performance results
- 23:29Future work and resources
- 24:57Q&A
related talks
49mSSDs for Your Database
25mFlashield: a Hybrid Key-value Cache that Controls Flash Write Amplification
43mCeph and Flash
40mFlashcache at mamba.ru
18mThe Solid State Storage Revolution
28mHow Can Databases Capitalize on Computational Storage?
20mMaximizing the Benefit of Flash Storage for Aerospike
20mExtreme Performance and SmartScaling MySQL with Storage Class Memory Servers