In some situations, DRS may not migrate VMs in a cluster to an empty host, even though it shows all signs of it being compatible.
There are no issues with moving VMs manually to the host.
vCenter Server 6.x
vCenter Server 7.x
vCenter Server 8.x
vCenter Server 9.x
DRS will not migrate VMs if the benefit outweighs the cost of the migration. Therefore, a cluster full of VMs with excellent DRS scores (90% +) will usually not create recommendations to move any VMs to an empty host. This can be perceived as an imbalance in the cluster.
You can confirm the DRS scores for VMs in the UI by clicking on the Cluster Object > Monitor > vSphere DRS > VM DRS Score.
Network issues can cause DRS to determine the cost of a migration to be higher than the benefit.
No action is needed.
The VMs will be moved when enough contention is created in the cluster to merit a DRS recommended vMotion. Additionally, newly powered on VMs will likely come up on the empty host.
If the DRS score is less than 90%, but VMs still are not migrated via DRS, check that all necessary uplinks are up.
If any uplinks are down, bring them up and check for DRS migrations. If the downed uplinks are not needed, consider removing them.
Even distribution of VMs:
If you want to have an even distribution of VMs across the hosts in the cluster, you can enable VM Distribution option. For details, refer to page 84 below:
https://www.vmware.com/docs/vsphere-esxi-vcenter-server-80U3-performance-best-practices
Changing the CPU Over-Commitment ratios:
If you need to strictly constrain the maximum vCPU to pCPU ratio allowed per host:
In the same vSphere DRS Edit Settings window, expand Additional Options.
Under CPU Over-Commitment, set either:
MaxVcpusPerClusterPct: Controls the cluster-wide vCPU:pCPU percentage (e.g., set to 100 for 1:1 ratio).
MaxVCPUsPerCore: Caps the maximum vCPU to physical core ratio allowed on any single ESXi host.
Click OK. Verification: Review host total vCPU allocations in the cluster summary tab to verify no host exceeds the enforced ratio threshold.
For details refer to: https://knowledge.broadcom.com/external/article/344239/the-total-number-of-virtual-cpus-present.html
If another host in the cluster is put into maintenance mode and VMs are not migrated to the empty host, there may be an issue with compatibility of the host. Check the vCenter /var/log/vmware/vpxd/vpxd.log with the logging level set to verbose for more information.