Dell VxRail Architecture Overview

Normal cluster operations

Normal cluster operations include the traffic between data nodes, VMware vCenter Server, and the witness appliance. Most traffic is between data nodes and VMware vCenter Server and is affected by the number of VMs and components, but is usually light. Measurements of a cluster with 25 VMs and nearly 1,000 components indicated a bandwidth consumption lower than 0.3 Mbps.

Witness appliance installation

The following are the witness appliance installation requirements:

• Install, configure, and add the witness appliance to the VMware vCenter Server inventory before deploying the VxRail 2-node cluster with VMware vSAN.
• For VxRail 7.0.410 and later, the witness appliance is added to the VMware vCenter Server on Day 1 which is used by the 2-node cluster.
• The witness VMware vSAN traffic must be on a different vLAN than the witness traffic of the data node.
• The witness appliance must be managed by the same VMware vCenter Server that is managing the 2-node cluster.
• VMware vSAN 2-node cluster requires WTS. WTS requires a subnet for the VMkernel port on the host to communicate with the witness appliance. For VxRail 7.0.100 and earlier versions, a witness appliance can only be connected to one VMware vSAN 2-node cluster.
• For VxRail 7.0.132 and later versions, a single witness can be shared with up to 64 2-node clusters.
  • The witness can run in the same physical site as the VMware vSAN data nodes. The witness cannot be placed in the 2-node cluster to which it provides quorum.
  • Place the VMware vSAN witness host in a different data center such as a main data center or a cloud provider.
• While possible, do not deploy the witness appliance on another 2-node cluster. Contact VMware Support for this solution design.

Witness contingencies

The following are the witness contingencies:

• The witness appliance does not maintain any data, only metadata components.
• The witness traffic can be influenced by the I/O workload running in the cluster. Generally, there is light traffic while the cluster is in a normal state.
• If the primary node fails or is partitioned, the following events occur:
  • VMware vSAN powers off the VMs in the failed host.
  • The secondary node is elected as the HA primary. The witness host sends updates to the new primary, followed by the acknowledgment from the primary that the ownership is updated. Each component update requires 1,138 bytes.
  • When the update is completed, a quorum is formed between the secondary host and the witness host. This allows VMs access to their data and be powered on.
• The failover procedure requires enough bandwidth to allow for the ownership of components to change within a short interval of time.
• For a 2-node cluster with up to 25 VMs, verify that at least 0.8 Mbps be available to ensure a successful failover operation.

Maintenance, LCM, and troubleshooting

The following are the maintenance, LCM, and troubleshooting considerations:

• The wanted transfer times for large files primarily determines the amount of bandwidth that is reserved for maintenance, life cycle management, and troubleshooting.
• The log files that are used in troubleshooting are compressed and can be transferred in a reasonable time.
• The composite files that are used for software and firmware upgrades can be up to 4.0 GB and can take a long time to be transferred through a T1 link. Evaluate bandwidth requirements if you have specific maintenance window requirements. If you are using a T1 link, at least one Mb/s of bandwidth should be available for the transfer of the composite file. This transfer may take about nine hours.