Benchmarking: Repeatable & Comparable
summary
Practical checklist for storage benchmarking: use fio only (other tools mislead), precondition block devices and pre-write or thick‑provision VMs, run long enough to force SSDs past their SLC cache filled after ~10GB burst so you measure sustained tri‑modal behavior (example: 1,550 MB/s for ~7s then ~450 MB/s sustained). Watch Ceph RBD thin provisioning (4 MB objects cause first‑touch I/O stalls, initial random write IOPS <200 then to ~1,400), avoid unaligned 4 KB penalties and verify PCIe/memory/network saturation.
AI-written summary · about
chapters (10)
- 0:00Introduction to storage benchmarking
- 1:03Repeatable and comparable benchmarks
- 2:04Hard disk performance variations
- 7:17SSD performance characteristics
- 13:14SSD garbage collection impact
- 18:08System component bottlenecks
- 26:14Read modify write overhead
- 31:30Sparse allocation pitfalls
- 35:26Benchmarking best practices
- 39:19Q&A and closing remarks
related talks
44mBenchmark testing
04mLightning Talk - Benchmarking Persistent Data Volumes
30mSSD and HDD Performance Testing
35mBenchmarking: You're Doing It Wrong
40mAchieving the Ultimate Performance with KVM
34mEvaluating the Performance and Power Efficiency of Ceph Storage Configurations
38mScaling IO-bound application in the cloud
06mIGNITE: How to benchmark … poorly