Optimizing Ceph Object Storage for Production in Multisite Clouds
summary
Sizing and tuning advice for Ceph RGW clusters with concrete knobs: choose media by workload (NVMe/SSD for index/DB, HDD for cold), size BlueStore DB ~4% of data, expect metadata overhead ~~5% on large clusters, plan 1–3 year growth and avoid overly dense disks to speed backfill. Uses iperf for network testing, rgw-admin sync status and admin sockets to measure fetch bytes/pool latency, and playbooks for multisite setup; note dynamic sharding not supported and multisite replication is pull‑based so recent writes (minutes to ~hours) can be lost during failover.
AI-written summary · about