Ceph Operations at CERN: Where Do We Go From Here?
summary
CERN runs roughly 20 PB of Ceph and reports concrete results: a 3 PB live migration with zero downtime, filestore→bluestore moves that cut CPU load and doubled read speed, RADOS gateway metadata jumping from ~2k to ~83k ops/s, and a CephFS entry on the IO-500 (rank 21). Use cases include a 5 PB tape buffer, 2 PB S3/Swift, all‑flash hyperconverged cells delivering ~5k 4k IOPS, and a Merkle‑tree backfill proof‑of‑concept to shrink recovery work. Balancer and Keystone sync solutions are in production.
AI-written summary · about
chapters (7)
- 0:00Introduction and Ceph history at CERN
- 5:03Recent operations and notable experiences
- 12:35BlueStore migration and performance gains
- 17:35HPC workloads and CephFS optimizations
- 20:43Backup systems and physics analysis demos
- 22:29Ongoing work and future plans
- 27:08Q&A session
related talks
18mKeynote: The Road to Open Source Ceph: Bloomberg’s Enterprise Ceph Journey
19m10 years of OpenStack at CERN (From 0 to 300k cores)
15mThe CERN Cloud Infrastructure: 5 years in 15 minutes
18mScaling Bare Metal Provisioning with Nova and Ironic at CERN
16mTips and Tricks for Managing and Administering Ceph Clusters
22mMoving a Large Workload from a Public Cloud to an OpenStack Private Cloud...
39mHops in the Cloud
20mKeynote Welcome