Configuring Ceph Deployments with an Easy to Use Calculator
summary
Empirical benchmarks on an all‑flash Ceph cluster (Broadwell servers, Intel NVMe 4500/Optane hardware, Bluestore, 3x replication, IO tests via CBT and Grafana graphs) drive the recommendations and an Excel template is provided to reproduce them. Measured metadata averages ≈20 KB per object, translating to roughly 5 GB RocksDB per OSD (far below the 4% doc rule that would yield ~1.7 TB on a 40 TB OSD). WAL sizing targets ~100 GB (≈80 s buffer) and the cluster sustains up to ~100 Gbps per node.
AI-written summary · about