Hardware and network questions

What kind of server chassis should I use for IBM Storage Ceph?

Begin by considering your default domain. Start with at least 4–7 server chassis, even if the chassis are if not fully populated. For denser environments, ensure that power, cooling, and floor load limits are evaluated. Overly dense HDD servers can create bottlenecks and stress physical infrastructure.

For more information, see Server and rack solutions.

What are the recommended hardware specifications for storage nodes?

IBM recommends two general server configurations. Either SSD-based nodes and HDD-based nodes. For more details about each server configuration type, see Table 1.

Table 1. Recommended hardware specifications
SSD-based nodes	HDD-based nodes
1U 10–14 NVMe SSDs 48–64 vCores 128 GB+ RAM 10–25 Gb/s bonded network	2U 12 or 24 LFF SATA HDDs IT-mode HBA 128–256 GB RAM 10–25 Gb/s bonded network

What kind of drives should I choose when using Ceph OSDs?

The following drives are recommended when using Ceph OSDs:

SSDs, especially NVMe, are preferred for high performance, particularly for RBD workloads.
HDDs are viable for archival/cold data. Use SSDs as metadata offload devices for improved performance.
Hybrid configurations (HDDs + SSDs) can optimize cost and performance.

Are RAID cards needed in Ceph?

No. Ceph does not benefit from RAID controllers. Use plain IT-mode Host Bus Adapter (HBA). Invest that budget into SSDs or faster networking instead.

How much CPU and memory do I need?

Table 2 breaks down the different of CPU and memory types and their recommended values.

Table 2. CPU and memory recommendations
CPU or memory type	Recommendation
HDD OSDs	Minimum 2 vCores per OSD
SSD OSDs	Recommended 4–6 vCores per OSD
Total vCores	48–64, for SSD-heavy nodes
RAM	Minimum 128 GB Maximum 256 GB, depending on workload

What are the network requirements for Ceph?

Following are the recommended network requirements:

IBM Storage Ceph requires 10 Gb/s per node.
25–100 Gb/s for SSD nodes or dense HDD configurations.
Bonding improves resilience. For example, Link Aggregation Control Protocol (LACP).
A dedicated replication/recovery network is optional but can help in 10 Gb/s setups.

For more information on network requirements, see Network Requirements in IBM Storage Ceph.

Should I bond network interfaces?

Yes, especially for resilience and maintenance flexibility. Bonding is common in small to mid-sized clusters.

Should I deploy a separate network for Ceph internal traffic?

Deploying a separate network for Ceph internal traffic is optional. Consider it if you're using 10 Gb/s NICs or high-density storage. For 25 Gb/s and higher setups with bonding, it may be unnecessary.