Understanding Ceph Cluster Planning

Designing a robust Ceph cluster requires careful balancing of raw storage capacity, data redundancy, and performance. Whether you are deploying a Hyper-Converged Infrastructure (HCI) with Proxmox VE or a dedicated storage cluster, understanding the metrics calculated above is essential for avoiding data loss and performance bottlenecks.

Raw vs. Usable Capacity

One of the most common confusion points for new administrators is the difference between Raw Capacity (the sum of all hard drives) and Usable Capacity (actual space for files).

• Replication (Default): In a standard 3x replication setup, every object is stored three times. This provides high availability and fast recovery but results in a 33% storage efficiency. For every 1 TB of usable data, you need 3 TB of raw storage.
• Erasure Coding (EC): For archival or cold data, EC is more efficient. It splits data into k data chunks and m coding chunks. For example, a 4+2 profile offers 66% efficiency (similar to RAID 6) but requires more CPU power for writes.

Placement Groups (PGs) Explained

Placement Groups (PGs) are internal data structures that map objects to OSDs (Object Storage Daemons). Getting the pg_num right is critical:

• Too few PGs: Leads to poor data distribution and uneven OSD usage.
• Too many PGs: Causes high CPU and RAM usage, potentially crashing OSDs during rebalancing.

The general rule of thumb used by this calculator is approximately 100 PGs per OSD. The value is always rounded to the nearest power of 2 to ensure CRUSH map stability.