Keeping vSphere Clusters Balanced Without Manual vMotions

You open the vSphere Client for a three-host cluster and see Host A at 85% CPU ready while Hosts B and C sit near 30%. The VMs are fine, but the imbalance is a risk. The useful takeaway is that vSphere Distributed Resource Scheduler with vMotion can automate rebalancing, but the automation level, migration threshold, and constraints decide whether it helps or creates churn.

What DRS actually evaluates

DRS continuously monitors CPU and memory demand across hosts in a cluster. It builds a placement model and recommends live migrations via vMotion so no single host becomes a hotspot. vMotion moves a running VM between hosts without downtime, provided the prerequisites are met: shared storage or supported vMotion without shared storage, a VMkernel port enabled for vMotion traffic, and CPU compatibility.

CPU compatibility is usually handled with Enhanced vMotion Compatibility, EVC. EVC masks newer CPU features so VMs can move between hosts with different generations. Enabling EVC later may require VM downtime, so it is a design decision made before VMs are powered on.

Automation level and migration threshold

DRS automation is set per cluster: Manual, Partially Automated, or Fully Automated. Manual only suggests moves. Partially Automated applies recommendations but prompts for confirmation. Fully Automated applies them without prompting and is the common production choice when operators trust the rules.

The migration threshold slider controls how much imbalance DRS tolerates before acting. Conservative tolerates more imbalance and moves less often. Aggressive tolerates little imbalance and moves more often. Aggressive settings with large VMs can increase vMotion events, adding network and CPU overhead and the risk of vMotion storms.

Rules that constrain placement

DRS does not balance storage. Storage DRS is a separate feature for datastore clusters. DRS rules let engineers express placement intent: VM-VM affinity/anti-affinity and VM-Host rules. For example, an anti-affinity rule can keep two domain controllers on separate hosts for fault isolation, while a must-run-on-host rule can pin a license-restricted workload.

Predictive DRS, available with vRealize/Aria Operations integration in supported versions, can rebalance ahead of forecasted demand rather than reacting after contention occurs. Feature availability and UI names vary across vSphere 6.7, 7.0, 8.0 and licensing editions. DRS requires appropriate vSphere licensing; it is not available on the free ESXi hypervisor.

Worked example: rebalancing a hot host

Consider a cluster with three hosts, Host A, Host B, Host C, with DRS enabled in Fully Automated mode and threshold set to Normal. Host A runs six CPU-intensive VMs and shows high CPU ready. Hosts B and C have spare capacity.

In vCenter, Cluster > Configure > vSphere DRS shows the automation level and threshold. Cluster > Monitor > DRS shows the current balance score and recent recommendations. DRS would evaluate the cluster, identify VMs on Host A that can be moved with minimal impact, and issue vMotion recommendations. With Fully Automated, eligible moves would be applied automatically.

Before relying on automatic moves, verify vMotion prerequisites. Attempt a manual vMotion between hosts and review any compatibility errors. If hosts have different CPU generations, configure an EVC baseline for the cluster at a level supported by all hosts. Confirm the VMkernel port for vMotion is configured on a dedicated network segment.

Trade-offs and limitations

DRS improves utilization but does not guarantee perfect balance. Aggressive thresholds can cause frequent migrations that add overhead without meaningful performance gain. Large VMs take longer to migrate and can temporarily saturate vMotion networks.

DRS also respects resource reservations and limits. A VM with a high reservation will not be moved to a host that cannot honor it. Rules can prevent the ideal placement, so review rules regularly.

Keep vCenter Server Appliance patched. Management interfaces have had security advisories over time, and an unpatched vCenter undermines automation reliability.

Actionable checks

Confirm your vSphere version and edition support DRS and the desired automation mode by checking the feature matrix in your environment's licensing and the vCenter UI.

In a non-production cluster, enable DRS in Manual mode first, load one host with CPU-intensive VMs, and observe recommendations in Monitor > DRS. Correlate vMotion events in Tasks & Events with VM performance metrics to confirm balancing helps rather than churns. Only then move to Partially Automated or Fully Automated with a threshold you can justify.

Review DRS migration threshold and VM rules quarterly, especially after hardware refreshes or workload changes.