Ceph Performance: The Basics of Tentacle Racing
summary
Argues that disciplined, deterministic benchmarking yields repeatable Ceph performance and shows concrete tooling and practices: CBT YAML workloads add deterministic prefill/preconditioning and support total IO depth for finer-grain curves; runs use 3–5 minutes per point and expect ~5–10% variance. Builds a new report generator with an intermediate format, PDF/Markdown outputs, comparison tables (e.g. throughput at 2 ms), 30+ hockey-stick curves, block-level support now and planned object/NVMe, currently two-step but moving toward single-click.
AI-written summary · about
chapters (9)
- 0:00Introduction to talk
- 1:04Why max IOPS isn't enough
- 3:55Preloading and steady state
- 6:50Benchmarking gotchas
- 10:27Introducing CBT
- 12:36CBT improvements
- 17:39SEPHADM support
- 19:48Future CBT features
- 22:41Conclusion and Q&A
related talks
05mCeph BenchOps: Following DevOps Practices for Benchmarking Ceph
28mA Welcoming Ceph Experience Improvements for First-Time Users or: How to Make Beginners Love Ceph
40mJuggling Petabytes: Managing Ceph at Scale with Ceph-ansible
17mKeynote: Ceph Journey - A Perspective
05mHealthier Ceph Clusters with Ceph-medic
36mRook: Intro and Deep Dive With Ceph Storage
44mContainer Management at Scale
29mReliable and Cost-Effective: Open Storage Strategies for Kubernetes