In a vSphere 6.5 cluster, the administrator notices that it is possible to set a percentage in the DRS option, "Control CPU over-commitment in the cluster." Which is the purpose of this setting?

The setting is based on a vCPU-to-pCPU count ratio, not a GHz consumption comparison, so it does not cap aggregate virtual machine CPU usage measured in GHz.

This setting controls the vCPU-to-pCPU ratio across the cluster by capping how many powered-on virtual machine vCPUs can exist relative to the total number of physical CPU threads available. When the ratio limit is reached, DRS will prevent additional virtual machines from powering on in the cluster. This directly governs CPU over-commitment rather than raw GHz throughput.

The setting limits VM power-on operations once the ratio threshold is met, but it does not prevent the creation (provisioning) of virtual machines in the cluster.

The setting does not involve shares or compare shares to available GHz; shares are a separate resource entitlement mechanism used during CPU contention scheduling.