Ceph Capacity Calculator

Explanation of Settings

To ensure your Ceph cluster planning is successful, it is important to understand the technical parameters used in this calculator:

• Unit Basis (TB vs. TiB): Hard drive manufacturers use decimal Terabytes (Base 1000). However, Ceph and operating systems calculate in binary Tebibytes (Base 1024). A “12 TB” drive provides only about 10.9 TiB of raw capacity in Ceph.
• Redundancy Mode: * Replication: Creates full copies of your data (Standard: 3x). Highly performant and simple, but expensive (33% efficiency).
  • Erasure Coding (EC): Breaks data into chunks ($k$) and adds parity chunks ($m$). Much higher efficiency (often > 60%), but requires more CPU resources.
• N-1 Reserve: In a healthy cluster, there must always be enough free space to allow the data from the largest node to be redistributed to the remaining servers in the event of a failure (Self-Healing).
• Safe Fill Ratio (80%): Ceph should never be filled to 100%. At 85% (“Nearfull”), performance drops drastically. At 95% (“Full”), Ceph stops all write operations to prevent data corruption.

---

What is Ceph?

Ceph is an open-source software-defined storage solution designed to run on standard hardware (commodity servers). Instead of buying a single, expensive storage filer, Ceph connects many individual servers (nodes) into one massive storage pool.

The core advantages of Ceph are:

1. No Single Point of Failure: If a server or a disk fails, the data remains available.
2. Unlimited Scalability: You can add new nodes at any time while the system is running.
3. Self-Healing: The system detects errors automatically and restores the desired redundancy without manual intervention.

---

Best Practices for Maximum Efficiency

Planning a cluster is always a balance between cost (efficiency), security, and performance.

How many nodes are required?

• Replication (3x): You need at least 3 nodes. With only 2 nodes, a single failure would leave you with no copy to “heal” the cluster back to health.
• Erasure Coding: The rule of thumb is $k + m + 1$ nodes. For an EC 4+2 profile (4 data chunks, 2 parity chunks), you should plan for at least 7 nodes. This way, one entire server can fail, and Ceph still has enough “targets” to redistribute the data in a $4+2$ scheme.

Efficiency Comparison

Strategy	Min. Nodes (Recommended)	Efficiency (Net/Raw)	Security
Replica 3x	3	33.3%	High (Fast recovery)
EC 2+1	4	66.7%	Medium (1 failure tolerated)
EC 4+2	7	66.7%	Very High (2 failures tolerated)
EC 8+3	12	72.7%	Extremely High

Tips for Building Your Cluster

1. Uniform OSD Sizes: Whenever possible, use hard drives of the same size. Ceph fills OSDs based on percentage. If you mix sizes (e.g., 4 TB and 12 TB), the small drives will fill up much faster, which can block the entire cluster.
2. Network Bandwidth: 10 GbE networking is the absolute minimum for Ceph. In the event of a disk failure, Ceph must copy terabytes of data across the network. A 1 GbE network would become an immediate bottleneck and cripple your application performance.
3. OSD Distribution: Aim for about 10 to 15 OSDs (disks) per node. Having too many disks in a single server increases risk: if that one server fails, the network must compensate for a massive amount of data all at once.