Basic and Advanced Analysis of Ceph Volume Backend Driver in Cinder
summary
Deep dive into the Cinder RBD backend showing concrete internals: volumes split into 4MB objects, placement groups map objects and default PG replication is 3x, and object data lives under PG-named directories; explains snapshot vs flatten vs transfer semantics and the export/import backup flow. Operational case: deployed across 8 Ceph clusters holding ~4 PB of data and added immutable image caching to eliminate OMAP recovery and greatly increase volume-from-image and instance provisioning rates.
AI-written summary · about
chapters (5)
- 0:00Introduction and speaker background
- 2:40Basic features of Cinder volume backend
- 10:04Snapshots and volume operations
- 17:49Advanced features and optimizations
- 23:30Multi-cluster replication and Q&A
related talks
01h 02mIntroduction to Ceph and Architectural Overview
24mThe Ceph Community: A Driving Force for Project Evolution
39mCeph Scientific Users Working Group BoF
13mKeynote: Getting to Know Ceph
32mDeploying Stand-alone Ceph and Integrating with K8/OpenShift and OpenStack
14mKeynote: Building and Sustaining Momentum in Open-Source Storage
36mTesting Ceph: Status, Development, & Opportunities
29mCeph, Xen, and CloudStack: Semper Melior