Tool · Cloud Computing
Ceph capacity calculator
How much space you really have left after replication or erasure coding, how many placement groups to set, and whether the cluster will survive a node failure. Calculated in your browser — nothing is sent to us.
Assumptions: capacity calculated in decimal TB; margin for the failure of one node; PGs rounded to a power of 2. The result is indicative — a production design must account for device classes, CRUSH maps and the workload profile.
Blue: usable · green: safe to fill
How to read the result
Three things that surprise people most often.
3× replication means 33% of capacity
Every object exists in three copies. 100 TB raw is ~33 TB usable. Erasure coding 4+2 gives 67%, but costs CPU and latency — for virtualization, replication is usually the choice.
Don't fill above 85%
Once full_ratio is exceeded, Ceph blocks writes. On top of that, the cluster needs room to rebuild data after a node failure — which is why safe capacity is noticeably lower than usable capacity.
With PGs, “more” is not “better”
Too few PGs means uneven data distribution; too many means memory pressure on OSDs. The target is about 100 PGs per OSD across all pools combined; the autoscaler in newer releases helps, but it pays to understand the numbers.