Status and Future of the Ceph File System
summary
Luminous stabilizes multiple active MDS so metadata load scales roughly linearly and you control actives with max_mds; an observed ping‑pong imbalance (experiment showed ~200,000 cached inodes against the 100k default) motivates subtree pinning via a directory xattr to force assignment and avoid balancer oscillation. Directory fragmentation removes the 100k readdir limit so directories can exceed 100,000 files. Cache accounting fixes move from ~65%→~80% measured usage and cut worst‑case RAM overuse from ~50% to ~25%. Mimic adds snapshots per directory, kernel quota (eventual enforcement), and an NFS‑Ganesha gateway managed by Kubernetes for HA and scale.
AI-written summary · about
chapters (6)
- 0:00Introduction to CephFS
- 4:54Features in Ceph Luminous
- 13:58Directory fragmentation in CephFS
- 18:16Upcoming features in Ceph Mimic
- 26:19NFS gateway integration plans
- 30:00Q&A session
related talks
34mRook Deployed Scalable NFS Clusters Exporting CephFS
44mContainer Management at Scale
27mThe Storage Architecture of Intel's Data Management Platform (DMP)
27mCeph-CSI: The Engine Behind Persistent Storage in Kubernetes
37mCache Me If You Can: P2P Image Sharing in Kubernetes with Spegel
27mSolving inefficiency challenges at Datacenter scale
31mRook: Intro and Deep Dive With Ceph Storage
40mCeph CSI Driver: Bridging Containers and Ceph