VMware recommends, where possible, using multiple server cabinets for the ESXi hosts, and distributing the vSphere clusters across the cabinets to minimize the impact of a single component failure.

This is also something to take into consideration when designing vSAN fault domains / rack awareness and vSphere HA admission control policies. If your tenant SLA dictates that you must maintain consumer services despite a blade chassis or compute cabinet failure, the number of hosts in the same failure domain that can be placed inside one of those entities is dictated by your admission control policy reservation. For instance, if the service provider guarantees to tenants that a blade chassis is not a single point of failure and the design has employed a Cisco UCS Blade System with 5108 blade chassis that are capable of supporting a maximum of 8 half-width blades, in a 24-node vSphere cluster, which can tolerate up to 3 hosts failing (13% of CPU and memory resource), each 5108 blade chassis must not accommodate more than 3 nodes that form part of the same vSphere cluster.

 

While a blade chassis or physical server rack is not typically considered a single point of failure, additional protection can provide higher levels of mitigation against major component outage.

The next consideration beyond chassis failure is mitigating against rack failure or even data center zone failure. The level of attention given to these requirements for availability and reducing the possibility of single points of failure must be stipulated in the design requirements.