Intro to Ceph, the Distributed Storage System
summary
Ceph description of components and concrete behaviors: a bottom-layer RADOS object store with CRUSH placement algorithm drives client-side placement so lookups avoid a central metadata server; thousands of OSDs and a small monitor quorum serve data and cluster maps; consistency uses peering and backfill recovery for placement-group failover; RBD adds snapshots, copy-on-write clones and RBD mirroring (asynchronous, crash-consistent, journal-backed); BlueStore replaces XFS for large speedups; Nautilus brings on-wire encryption, a manager, balancer and dashboard; CERN runs ~10,000 OSDs / 55 PB.
AI-written summary · about
chapters (32)
- 0:00Introduction
- 2:14Components of Ceph
- 3:00Metadata Server
- 4:04Crush
- 5:47Pools
- 19:50Mirroring
- 20:20Use Cases
- 20:41RVD Mirror Mirror
- 21:50Southwest
- 22:47SetFS
- 23:33Hard Links
- 24:55Metadata
- 25:44Recursive Statistics
- 26:11Security Model
- 27:59Standby Servers
- 29:33MVS Servers
- 30:07Cooperative Partitioning
- 30:37Balancing
- 31:36Directory Fragmentation
- 32:26Scrub Repair
- 33:34Parallel Workers
- 34:50Repair Tools
- 35:07Snapshots
- 36:58Multiple File Systems
- 38:04Pain Points
- 39:36Debugging
- 40:40Who should use FS
- 41:51Erase your coding
- 43:09Nautilus
- 44:34Web Dashboard
- 44:56Mimic
- 46:03Audience Questions
related talks
02h 59mCeph Storage: a Caffeinated Primer
29mCeph Operations at Scale
34mModernizing Ceph Deployments at CERN: CephFS and Object Storage Across Data Centres
35mMaking Ceph Fast in the Face of Failure
29mGlobal Deduplication for Ceph
12mCeph QoS: How to support QoS in distributed storage system
30mScaling Databases and File APIs with Programmable Ceph Object Storage
21mNew Techniques to Improve Small I/O Workloads in Distributed File Systems