Rapid Design and Effective Operating of a General Purpose Object Storage at RWTH
summary
Collaborative Ceph deployment with four partners builds a geo-redundant S3 system across 4 geo-distributed data centers over dark fiber, adding NAS and evaluating an iSCSI gateway and Windows fileservers; design locks in rack/rack-group placement early and limits single-node capacity to ~9% of OSDs. Uses erasure coding for cost efficiency but keeps small metadata/index pools replicated, prefers BlueStore over Filestore, automates config and patching via Ansible/CI, regularly reprovisions nodes instead of patching, splits gateway network, and centralizes logs for security and troubleshooting.
AI-written summary · about
chapters (6)
- 0:00Start
- 0:21Resiliency Use multiple datacenters Dark fiber/low latency is needed • Use appropriate levels in your crush hierarchy • Take failure domains into account A single node failure should not impact more than 9% of your cluster
- 1:40Efficiency . Use erasure coding Bluestore is more 10 efficient than Filestore What components of the nodes shall be redundant?
- 2:51Compliance and workload reduction Use Ansible • Automate update deployment Have a test environment Keep it up to date • Use a version control system for infrastructure code Do not fix installations, redeploy broken stuff
- 4:28Security • Automated patching speeds up remediation • Split gateway and public network Do security test Mozila observatory - Monitor TLS certificates Collect and combine logs
- 6:32Why where fast? Ceph is ready for production • Ceph is scalable -If you choose a "medium size", you can scale Default configuration is great
related talks
29mReliable and Cost-Effective: Open Storage Strategies for Kubernetes
36mStorage Wars
45mOpen Storage in the Enterprise with GlusterFS
20mEnterprise-Ready Ceph: Simplifying Storage Management, Operations, and Deployment
44mContainer Management at Scale
10mCephBox - A Home Cloud Personal Storage Solution
44mA Custom File Store for 350 PB of Video Content
31mScaling Autonomous-Driving R&D: A 100 PB, 100 Billion-File Open-Source Elastic File...