vSAN 6.x Cluster designs and considerations

In this post, we are going to discuss possible vSAN cluster designs, vSAN requirements and considerations.
As we all know, vSAN is available to customers in two flavours as listed below.
  • Hybrid: Hybrid vSAN is implemented using combination of SSD (as Cache) and HDD (as Capacity device)
Image: VMware
  • All Flash: All flash uses only SSDs (for Cache as well as Capacity device) for implementing vSAN.
Image: VMware
Hybrid vSAN was inducted in vSphere 5.5 onwards whereas All-Flash was inducted in vSphere 6.0 onwards.
vSAN requirements:
  • Min 3 ESXi Hosts (max 64 hosts)
  • Min 1GB network for Hybrid and Min 10GB network for All flash
  • RAID controller: Passthrough Mode is recommended, RAID 0 mode is also supported.
  • vSAN License
  • Storage device requirements for vSAN:
    • Disks to be used for vSAN must be raw.
    • While enabling vSAN cluster, first 3 ESXi hosts must have local storage.  Later you can add diskless servers as well. 
    • Min 1 disk group on first 3 ESXi host
      • Min and Max 1 SSD for caching per disk group
      • Min 1 HDD/SSD for Capacity device and Max 7 per disk group
    • Max 5 disk groups per ESXi

There are four different ways in which vSAN cluster can be designed. Below are the possible types for vSAN cluster.

  • Standard Cluster
  • Two Node cluster
  • Stretched Cluster
  • Two Node Stretched Cluster

Standard Cluster:

A standard vSAN cluster consists of a minimum of three physical nodes and can be scaled up to 64 nodes. All the hosts in a standard cluster are commonly located at a single location and are well-connected on the same Layer-2 network. Though maximum number of servers in vSAN cluster can be 64, maximum 32 fault domains can exist in a standard vSAN cluster.


Two Node Cluster:

Though we refer it as two node cluster, still there will be 3 hosts in this design to form a valid cluster. 

Two physical ESXi hosts are placed in same location and are used for hosting workloads. These hosts are usually connected to the same network switch or can be directly connected. While 10Gbps connections may be directly connected, 1Gbps connections will require a crossover cable.

A third ESXi host is required for a 2-node configuration to avoid “split-brain” issues when network connectivity is lost between the two physical nodes. Hence it will be used as witness host. Witness host is generally placed in different location than the Two nodes used for workloads. I will explain requirements for witness host later in this article.

Stretched Cluster:

A vSAN Stretched Cluster provides resiliency against the loss of an entire site. This design uses 3 sites to achieve site resiliency. Out of 3 sites, two sites are designated as Data sites (One site configured as preferred site and other as secondary site) where hosts are distributed evenly across two sites. Third site (referred as witness site) is used for placement of witness host and only used for hosting witness components. In stretched cluster design, there are only 3 fault domains since it’s a three site deployment.

The two sites are well-connected (Layer 2 stretched network) from a network perspective with a round trip time (RTT) latency of no more than 5 ms. Connectivity between data site and Witness site can be Layer 3 with latency requirement as mentioned below.

  • If there are 10 or more hosts per data site, latency must be 100 ms or less
  • If there are 10 or less hosts per data site, latency must be 200 ms or less.

A vSAN Witness Host is placed at a third site (Witness site) to avoid “split-brain” issues if connectivity is lost between the two data sites.

A vSAN Stretched Cluster may have a maximum of 31 hosts in the cluster i.e. 15 hosts max per data site so total 30 hosts across two data sites and 1 Witness host in witness site. In cases where there is a need for more hosts across sites, additional vSAN Stretched Clusters can be used.

Two node stretched cluster:

A two-node stretched cluster is effectively the same as a two-node cluster but the two vSAN nodes are geographically disparate. 

Two-node stretched cluster configuration steps are same as that of stretched cluster. You must designate one site as the preferred site, and the other site becomes a secondary or non-preferred site. The only key difference is that each site contains only one host.


Witness Host considerations:
Witness host stores metadata commonly called as “witness components” for vSAN objects. Virtual machine data such as virtual disks and virtual machine configuration files are not stored on the vSAN Witness Host. The purpose of the vSAN Witness Host is to serve as a “tie-breaker” in cases where data sites are network isolated or disconnected.

A vSAN Witness Host may be a physical vSphere host, or a VMware provided virtual appliance, which can be easily deployed from an OVA. When using a physical host as a vSAN Witness Host, additional licensing is required, and the host must meet some general configuration requirements.  

When using a vSAN Witness Appliance as the vSAN Witness Host, it can easily reside on existing vSphere infrastructure, with no additional need for licensing.

When using 2 Node clusters for deployments such as remote office branch office (ROBO) locations, it is a common practice for vSAN Witness Appliances to reside at a primary datacenter. It is possible to run witness host at the same ROBO site but would require additional infrastructure at the ROBO site.

vSAN Witness Hosts providing quorum for Stretched Clusters may only be located in a tertiary site that is independent of the Preferred and Secondary Stretched Cluster sites.

One vSAN Witness Host is required for each 2 Node or Stretched Cluster vSAN deployment.

Using the VMware provided vSAN Witness Appliance is generally recommended as a better option for the vSAN Witness Host than using a physical vSphere host. The utilization of a vSAN Witness Appliance is relatively low during normal operations. It is not until a failover process occurs that a vSAN Witness Host will have any significant utilization. 

When using a vSAN Witness Appliance, it is patched in the same fashion as any other ESXi host. It is the last host updated when performing 2 Node and Stretched Cluster upgrades and should not be backed up. Should it become corrupted or deleted, it should be redeployed. vSAN 6.6 introduced a quick and easy wizard to change the associated vSAN Witness Host.

Leave a Reply