A VMware vSAN 2-node cluster on VxRail consists of a cluster with two VxRail E-Series nodes, and a Witness host deployed as a virtual appliance. The VxRail cluster is deployed and managed by VxRail Manager and VMware vCenter Server.
A vSAN 2-node configuration is very similar to a stretched-cluster configuration. The Witness host is the component that provides quorum for the two data nodes if there is a failure. As in a stretched-cluster configuration, the requirement for one Witness per cluster still applies.
Unlike a stretched cluster, the vCenter Server and the Witness host are typically located in a main data center, as shown in Figure 1, and the two vSAN data nodes are in a remote location. Even though the Witness host can be deployed at the same site as the data nodes, the most common deployment for multiple 2-node clusters is to have multiple Witnesses hosted in the same management cluster as the vCenter Server. This deployment optimizes the infrastructure cost by sharing the vSphere licenses and the management hosts.
This design is facilitated by the low bandwidth that is required for the communication between data nodes and the Witness.
Figure 1. Design of 2-node cluster with direct connect
A vSAN 2-node configuration maintains the same high availability characteristics as a regular cluster. Each physical node is configured as a vSAN fault domain. This means that the virtual machines can have one copy of data on each fault domain. If a node or a device fails, the virtual machine remains accessible through the alternate replica and Witness components.
When the failed node is restored, the distributed resource scheduler (DRS) automatically rebalances the virtual machines between the two nodes. DRS is highly recommended and requires a vSphere Enterprise edition license or higher.