Choosing vSphere HA Percentage-Based Admission Control for Predictable Failover

Problem: Power‑on blocked after a host failure

You have a vSphere cluster that runs a mix of small and medium VMs. After a host fails, some VMs fail to restart because HA admission control prevented them from being powered on earlier, leaving you with stranded capacity and surprised users.

Thesis: Percentage‑based admission control gives a simple, flexible way to reserve just enough resources for the configured number of host failures while avoiding the fragmentation pitfalls of slot‑based policies.

How the percentage policy works

HA admission control is a guardrail that evaluates whether powering on a VM would leave the cluster unable to restart all protected VMs after the specified number of host failures. With the percentage policy you set two values:

• CPU reserve – a percentage of the cluster’s total CPU capacity that HA holds back.
• Memory reserve – a percentage of the cluster’s total memory capacity that HA holds back; VM overhead is included in this calculation.

The usable capacity for VMs is the remainder after subtracting those reserves. If a VM’s reservation would push the used CPU or memory above the usable limit, the power‑on request is denied.

Because the reservation is expressed as a fraction of the *current* cluster totals, the policy automatically adapts when you add or remove hosts, and it does not suffer from the slot‑inflation problem where a single large‑memory VM inflates every slot.

Worked example: sizing a two‑host cluster for one‑host failure

Suppose you have a cluster with two ESXi hosts:

• Host A: 20 GHz CPU, 128 GB RAM
• Host B: 18 GHz CPU, 96 GB RAM

Total cluster capacity: 38 GHz CPU, 224 GB RAM.

To survive the loss of one host you need to reserve roughly the capacity of the smaller host (the worst‑case loss). A quick way is to reserve 50 % of CPU and 50 % of memory, which gives:

• CPU reserve: 0.50 × 38 = 19 GHz
• Memory reserve: 0.50 × 224 = 112 GB

In the vSphere Client:

1. Select the cluster → Configure → vSphere Availability → Admission Control.
2. Choose Percentage of cluster resources as the policy.
3. Enter 50 for CPU and 50 for Memory.
4. Click OK.

HA now shows a usable capacity of 19 GHz CPU and 112 GB RAM for VMs. If you try to power on a VM that needs 20 GHz CPU (or 120 GB RAM) the operation is blocked, because it would exceed the reserved pool.

You can verify the calculation by looking at the Failover Capacity read‑out on the same screen; it should display the reserved percentages and the resulting available resources.

Trade‑off: availability versus utilization

Increasing the reserved percentages improves the cluster’s ability to survive host failures, but it also strands resources that could otherwise run VMs. For example, raising the memory reserve from 50 % to 60 % in the example above would hold back 134 GB RAM, leaving only 90 GB for workloads—a 20 % reduction in usable memory.

Conversely, setting the percentages too low risks that, after a real host failure, the remaining hosts cannot accommodate all VMs, leading to failed restarts or performance degradation. Disabling admission control entirely reclaims the full capacity but removes the guarantee that HA will have enough resources to restart VMs.

The sweet spot is usually found by:

1. Identifying the maximum host failure count you want to tolerate (most designs use 1).
2. Calculating the CPU and memory of the smallest host in the cluster (or using a weighted average if hosts are highly imbalanced).
3. Setting the percentages to roughly that fraction of total cluster capacity.
4. Validating with a test: fill the cluster with VMs until admission control blocks further power‑ons, then place one host in maintenance mode and confirm that VMs restart successfully.

Practical verification steps

1. In the vSphere Client, navigate to Cluster → Configure → vSphere Availability → Admission Control and note the displayed CPU and memory reservation percentages. 2. Use the cluster’s Summary tab to see the total CPU and memory.
Compute expected usable capacity: total × (100 % − reserve %) and compare with the Available Resources figures shown by HA. 3. To test, power on VMs until the Failover Capacity indicator turns yellow or red, indicating that further VMs would breach the reserve. Then evacuate a host (Enter Maintenance Mode) and observe whether VMs restart on the remaining hosts without manual intervention.

These steps let you confirm that the percentage policy behaves as expected without needing to guess slot sizes or manually adjust advanced settings.

Actionable closing

For most general‑purpose vSphere clusters, start with the percentage‑based admission control policy, reserve roughly the capacity of your smallest host (or the host you are willing to lose), and validate with a simple power‑on‑until‑blocked test. Monitor the usable capacity over time as you add or remove hosts or change VM reservations, and adjust the percentages only when your failure‑tolerance requirements change. This approach gives you a clear, predictable guardrail that avoids the fragmentation headaches of slot‑based sizing while keeping the cluster’s utilization visible and controllable.