Short answer: in vSAN ESA, RAID-5 works on as few as 3 hosts, where it uses a 2+1 layout (1.5x capacity). The more efficient 4+1 layout (1.25x) needs 6 hosts, and vSAN picks between the two automatically. RAID-6 (4+2) also needs 6 hosts. With Auto-Policy Management enabled, vSAN only picks RAID-5 as the default policy from 4 hosts. Plan for at least one host more than the minimum, so a failed host can be rebuilt.
vSAN ESA minimum hosts per RAID level
| Protection | Failures to tolerate | Minimum hosts (fault domains) | Capacity overhead |
|---|---|---|---|
| RAID-1 mirror | 1 | 3 | 2x |
| RAID-5 (2+1) | 1 | 3 | 1.5x |
| RAID-5 (4+1) | 1 | 6 | 1.25x |
| RAID-1 mirror | 2 | 5 | 3x |
| RAID-6 (4+2) | 2 | 6 | 1.5x |
| RAID-1 mirror | 3 | 7 | 4x |
Numbers from Broadcom TechDocs: Selecting the Best RAID Configuration for a vSAN Storage Cluster. In a stretched cluster, RAID-5 (2+1) with site mirroring needs 3+3 data hosts plus a witness.
How adaptive RAID-5 chooses 2+1 or 4+1
ESA has no 3+1 layout like OSA. It offers a single RAID-5 policy rule, and vSAN chooses the stripe width from the cluster size. The aim is to keep a spare host whenever possible, so a host failure can be rebuilt without leaving the object degraded for long:
| Hosts in cluster | RAID-5 layout | Spare host for rebuild |
|---|---|---|
| 3 | 2+1 | No |
| 4–5 | 2+1 | Yes |
| 6 or more | 4+1 | Yes |
If a 6-host cluster drops to 5 hosts, vSAN keeps the 4+1 objects for 24 hours. After that it reconfigures them to 2+1. When the cluster grows back to 6 or more hosts, vSAN moves RAID-5 objects to 4+1. Broadcom explains the mechanism in Adaptive RAID-5 Erasure Coding with the Express Storage Architecture. Either change moves data, so account for the extra capacity and I/O when you remove hosts from a 6-node cluster for a longer period.
Auto-Policy Management changes the thresholds
With Auto-Policy Management on an ESA cluster, vSAN creates the default datastore policy for you. The TechDocs page What are vSAN Default Storage Policies sets these rules:
- RAID-5 as the auto policy needs 4 hosts, even though RAID-5 itself works on 3.
- RAID-6 as the auto policy needs 6 hosts.
On a 3-host ESA cluster you can still assign RAID-5 with a manual policy. Be aware of the trade-off: during a host failure or long maintenance, you have no host to rebuild to.
Sizing guidance: absolute minimum vs practical minimum
- 3 hosts: RAID-1 or RAID-5 2+1 with FTT=1. It works, but a host outage leaves data unprotected until the host is back.
- 4–5 hosts: RAID-5 2+1 with a spare host. This is the common choice for small ESA clusters.
- 6 hosts: RAID-5 4+1 (best efficiency) or RAID-6 4+2 for FTT=2. RAID-6 on exactly 6 hosts has no spare.
- 7+ hosts: RAID-6 with a rebuild target.
Fault domains count, not hosts. If you configure rack fault domains, apply the table to the number of fault domains. Before buying hardware, also check the Broadcom Compatibility Guide for ESA-certified hardware. For vSAN 9 licensing and edition context, see the vSphere 9.0 product line comparison.
Further reading
- Duncan Epping: vSAN ESA and the minimum number of hosts with RAID-1/5/6
- Broadcom TechDocs: Selecting the Best RAID Configuration for a vSAN Storage Cluster
- Broadcom TechDocs: What are vSAN Default Storage Policies
