Check vSphere HA admission control: how much capacity a percentage or N+1 slot policy reserves and whether restarts actually fit.
Admission control reserves capacity in one of two shapes: a flat percentage of cluster CPU and memory, or slots sized so that losing N hosts still leaves room. Both reduce to the same test — cluster capacity minus current demand must exceed the demand of the VMs that were running on the hosts you are prepared to lose, checked separately for CPU and for memory. HA that cannot restart a host's worth of VMs is a false sense of safety, and the failure is silent until the day a host dies; equally, an over-generous reserve blocks legitimate power-on operations with an insufficient-resources error.
VMware HA Capacity
reserve = failover hosts ÷ total hosts; restart demand = ceil(VMs ÷ hosts) × failover hosts × average VM demand, which must fit in cluster capacity minus current demand.
reserve = failover hosts ÷ total hosts; restart demand = ceil(VMs ÷ hosts) × failover hosts × average VM demand, which must fit in cluster capacity minus current demand. Admission control reserves capacity in one of two shapes: a flat percentage of cluster CPU and memory, or slots sized so that losing N hosts still leaves room. Both reduce to the same test — cluster capacity minus current demand must exceed the demand of the VMs that were running on the hosts you are prepared to lose, checked separately for CPU and for memory.
HA that cannot restart a host's worth of VMs is a false sense of safety, and the failure is silent until the day a host dies; equally, an over-generous reserve blocks legitimate power-on operations with an insufficient-resources error.
This calculator takes 9 inputs: Hosts in cluster, CPU capacity per host, Memory per host, ESXi host memory overhead, Powered-on VMs, Average CPU demand per VM, Average memory per VM, Admission control policy, Percentage reserved (percentage policy). The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
With mixed hardware, a percentage reserve calculated from the cluster average is too small when the biggest host fails. Size the reserve so the single largest host's workload fits, then convert that back into a percentage.