Academy/VCP-VCF 9.0 Architect (2V0-13.25)/HA Admission Control Validation
This lab targets VCF 9.0

HA Admission Control Validation

VCF 9.0Intermediatevcp-foundation⏱ 75 min

Objectives

  • Understand HA admission control policies (slot-based and percentage-based)
  • Calculate slot sizes for different VM profiles in a 4-host cluster
  • Evaluate admission control scenarios and policy recommendations
  • Design N+1 HA policy for VCF management domain

Prerequisites

VCF 9.0 lab environment deployed with 4-host management cluster

Prior labs: vcp-architect-01: Cluster Basics

Required skills:

  • Understanding vSphere HA concepts
  • Basic cluster architecture knowledge

Lab Environment

4-host VCF management cluster (each host: 96GB RAM, 16 vCPU cores, 2.5GHz)

Cluster specification

ItemValue
hosts4
host ram gb96
host cores16
host frequency ghz2.5

Tasks

Task 1 Understand HA Admission Control Policies

Demonstrates ability to analyze admission control mechanisms and understand trade-offs between predictability (slot-based) and flexibility (percentage-based).

Establish foundational knowledge of slot-based and percentage-based admission control mechanisms and their impact on cluster capacity planning.

Step 1

Review the 4-host VCF management cluster. From each host's Summary tab record the CPU and memory available to VMs (the host's root resource pool, which already excludes the resources used for virtualization): in this lab assume 36,000 MHz and 88 GB per host, so 144,000 MHz and 352 GB for the cluster. vSphere HA uses these root-resource-pool figures, not the physical totals, and considers only connected hosts that are not in maintenance mode and have no HA errors.

Cluster capacity summary: 4 hosts, 36,000 MHz and 88 GB available per host (root resource pool), 144,000 MHz and 352 GB in total.
Do not confuse hypervisor reserved memory with VM memory reservations—they are calculated separately.
HA counts the host root resource pool, which excludes virtualization overhead. Read it from the host Summary tab; for identical hosts, divide the cluster figures by the host count.
Step 2

Define three VM profiles with reservations, because the slot policy works on reservations, not on configured vCPU or vRAM: Small 2 vCPU/4 GB with a 500 MHz and 4 GB reservation; Medium 4 vCPU/16 GB with 2,000 MHz and 16 GB; Large 8 vCPU/64 GB with 4,000 MHz and 32 GB (memory overhead for the Large VM about 0.4 GB). Slot size (vSphere 9.0 Availability, Slot Policy Admission Control): CPU component = the largest CPU reservation of any powered-on VM (32 MHz if none is set; das.vmcpuminmhz changes the default); memory component = the largest memory reservation plus memory overhead of any powered-on VM (no default). Compute the slot size for a cluster where at least one VM of each profile is powered on.

Slot size = 4,000 MHz CPU (largest CPU reservation, from the Large profile) and 32.4 GB memory (largest memory reservation 32 GB plus about 0.4 GB overhead, Large profile). One slot size applies to the whole cluster; every powered-on VM consumes one slot regardless of its own size, which is why a single large reservation distorts the calculation.
Configured vCPU and vRAM play no part in the slot size. A VM with no reservation contributes the 32 MHz default (changeable with das.vmcpuminmhz; vSphere 9.0 docs: Slot Policy Admission Control) and only its memory overhead; a VM with a 32 GB reservation sets a 32 GB slot for everyone.
Slot size is a cluster-wide pair of numbers in MHz and MB, not a fraction of a host. Cap a distorting component with das.slotcpuinmhz or das.slotmeminmb (VMs larger than the capped slot then need several slots; the vSphere Client shows how many VMs require multiple slots).
Step 3

Determine the slots per host: divide the host's available CPU by the CPU slot component and round down; do the same for memory; the smaller result is the host's slot count. Here: floor(36,000 / 4,000) = 9 and floor(88 / 32.4) = 2, so 2 slots per host and 8 slots in the 4-host cluster. Record how many VMs are powered on (for this lab: 6).

Slots per host: min(9, 2) = 2. Cluster total: 8 slots. Memory is the limiting component because of the Large reservation.
Do not divide the cluster total by the slot size: 352 / 32.4 = 10.9 is wrong, because the per-host rounding (2 per host) applies first.
Each host's count is computed separately and rounded down before summing; unequal hosts have unequal slot counts.
Step 4

Compute the Current Failover Capacity for Host Failures Cluster Tolerates = 1: remove hosts starting from the largest and check whether the remaining slots still satisfy every powered-on VM. With 6 powered-on VMs and 8 slots: one host fails, 6 slots remain, 6 VMs fit, so capacity = 1; two hosts fail, 4 slots remain, insufficient. Admission control therefore allows 8 - 2 - 6 = 0 further power-ons; a seventh VM is refused.

Current Failover Capacity = 1 host (6 slots remain after the largest host fails, enough for 6 powered-on VMs). Available slots for new power-ons = 0. Configured Failover Capacity = 1 is met; the cluster is at its admission limit although 352 GB of memory is mostly idle, a consequence of the 32.4 GB slot.
Ignoring N+1 means a host failure finds no room to restart its VMs. Admission control refuses power-ons that would break the configured failover capacity; disabling it to force a power-on removes that protection.
N+1 in slot terms means the slots of the largest host are held back. If the real reservations differ widely, consider das.slotmeminmb or the Cluster resource Percentage policy (Task 2).

Validation Gate

Check: Verify admission control calculations and N+1 capacity planning

Expected: Slot size in MHz and MB from the profile reservations, slots per host and per cluster, Current Failover Capacity for Host Failures Cluster Tolerates = 1, and the available-slot count for new power-ons

Common Errors

Slot size from configured vCPU/vRAM or as a host fraction
Symptom: Slot sizes written as 0.125 or 0.727 'of a host', or based on 4 GB/16 GB/64 GB configured memory
Cause: The slot policy uses reservations: largest CPU reservation (32 MHz default) and largest memory reservation plus overhead, in MHz and MB
Fix: Set reservations per profile, take the largest CPU and the largest memory reservation across powered-on VMs, and express the slot in MHz and MB
Dividing cluster totals by the slot size
Symptom: '352 GB / 32.4 GB = 10 slots'
Cause: Slots are counted per host (floor of CPU and of memory, then the smaller) and summed
Fix: Compute each host's slots from its root-resource-pool capacity, round down, take the smaller of CPU and memory, then add the hosts
Reading the failover capacity as 'one spare host'
Symptom: Treating N+1 as 5 hosts' worth of resources or as 3 hosts' worth divided by a slot fraction
Cause: Current Failover Capacity is the number of hosts, largest first, that can fail while the remaining slots still cover every powered-on VM
Fix: Remove the largest host's slots, compare the remainder with the powered-on VM count; repeat for the next host; the count of successful removals is the capacity

Task 2 Evaluate Admission Control Scenarios

Demonstrates ability to evaluate trade-offs between admission control methods and make data-driven policy recommendations backed by concrete calculations.

Apply admission control calculations to real-world scenarios and compare policy approaches (slot-based vs. percentage-based) for cluster design decisions.

Step 1

Scenario 1: a new VM with a 96 GB memory reservation (plus about 0.6 GB overhead) is requested for the management cluster. Recompute the slot policy from Task 1: the memory slot component becomes 96.6 GB, so each host supports floor(88 / 96.6) = 0 slots and the cluster has 0 slots; admission control refuses the power-on (and would also flag the existing VMs as unprotected). Determine the alternatives: lower the reservation, cap the slot with das.slotmeminmb (the 96 GB VM then needs several slots), or switch to the Cluster resource Percentage policy.

Determination: the 96 GB-reservation VM cannot be admitted under the slot policy; with its reservation it drives the slot to 96.6 GB and no host can hold a slot. Options recorded: reduce the reservation, das.slotmeminmb cap with multi-slot VMs, or percentage-based admission control.
Attempting to force-admit an oversized VM by disabling HA is a critical production risk. The VM would be unprotected.
A single VM reservation larger than any host's available memory makes slot admission control impossible; even a reservation of 48 GB would halve the cluster to 4 slots.
Step 2

Calculate the impact of losing one host. From Task 1: 4 hosts, 2 slots each, 8 slots, 6 VMs powered on, Current Failover Capacity 1. After one host fails: 3 hosts, 6 slots; all 6 admitted VMs restart (each takes one slot). Now test the oversubscribed case: if admission control had been disabled and 7 VMs were powered on, 6 slots remain for 7 VMs and one VM cannot be restarted with its reservation honoured.

With one host failed: 6 slots remain on 3 hosts; 6 admitted VMs fit (Current Failover Capacity 1 confirmed). With 7 VMs powered on against admission control, one VM has no slot after the failure.
Powering on beyond the slot count (by disabling admission control) causes restart failures when a host fails: the excess VMs stay down.
During failover HA restarts the failed host's VMs on the survivors; verify the powered-on VM count against the slots of (n-1) hosts, largest host removed first.
Step 3

Compare slot-based vs. percentage-based admission control for this cluster. Slot policy: one cluster-wide slot (here 4,000 MHz / 32.4 GB set by the Large reservation), 2 slots per host, 8 slots, 6 VMs admitted; hard limit, predictable, but wasteful when reservations differ widely. Percentage-based: Reserves a percentage (e.g., 25%) of cluster resources for failover; allows heterogeneous workloads without recalculation. For this cluster: the slot policy holds back the largest host's 2 slots (25% of the slots) and strands most of the 352 GB; Cluster resource Percentage with 25% CPU and 25% memory reserved admits VMs until their reservations reach 75% of the cluster's resources, so mixed reservations (vCenter, NSX, SDDC Manager, VCF Operations) fit without a common slot. Recommendation: Compare predictability vs. flexibility trade-off.

Expected

Comparison table:

  • Slot-based: Predictable, hard limits, requires slot recalculation for new VM types, prevents oversubscription
  • Percentage-based: Flexible, accommodates heterogeneous workloads, simpler to understand, less granular control

For this VCF management cluster the appliances carry different reservations (vCenter, NSX Manager nodes, SDDC Manager, VCF Operations), so the slot policy is distorted by the largest one; Cluster resource Percentage (25%) is the practical choice, or a slot policy with a das.slotmeminmb cap.

Percentage-based with insufficient reserve percentage (e.g., 10%) may fail to protect oversized VMs during failover.
Percentage-based is often easier to communicate to stakeholders: 'We reserve 25% capacity for failover' is simpler than explaining slots.
Step 4

Formulate a policy recommendation for this VCF management cluster. Decision: Which admission control method and why? Justification: 'Recommend Cluster resource Percentage admission control (25% CPU, 25% memory, equal to one of four hosts) for this cluster because: (1) the management appliances carry different reservations, so a single slot set by the largest one strands capacity; (2) the percentage reserve still guarantees one host failure; (3) new appliance sizes need no slot recalculation; (4) if deterministic slot accounting is required, use the slot policy with das.slotmeminmb capped at the typical appliance reservation and accept multi-slot VMs.' Document the policy choice with rationale.

Written recommendation: 'Implement Cluster resource Percentage admission control reserving 25% CPU and 25% memory (one host of four) for N+1 failover, with Host Failures Cluster Tolerates = 1 documented as the equivalent slot-policy setting. Admission is then governed by the sum of VM reservations against 75% of the cluster, not by a 32.4 GB slot that limits the cluster to 8 VMs.'
Never recommend a policy without understanding the specific workload. 'Use percentage-based for flexibility' without knowing the workload requirements is insufficient justification.
Always justify policy recommendations with specific cluster characteristics, workload profiles, and business requirements (e.g., 'VCF management domain requires N+1 availability for 99.95% uptime SLA').

Validation Gate

Check: Verify scenario calculations and policy recommendation

Expected: Documented capacity impact analysis, policy comparison, and justified recommendation

Common Errors

Assuming N+1 Means 'Have 1 Extra Host'
Symptom: Calculating capacity as if the cluster has 5 hosts worth of resources (4 hosts + 1 extra)
Cause: Misinterpreting N+1 as adding a host rather than reserving one host's worth of capacity
Fix: N+1 means capacity of (n-1) hosts is available for VMs; 1 host worth of capacity is reserved for failover. For 4 hosts: 3 hosts = available capacity, 1 host = reserved.
Not Considering VM Profile When Recommending Policy
Symptom: Recommending percentage-based for a cluster with homogeneous VM sizes, losing precision
Cause: Not analyzing the actual VMs in the cluster before choosing admission control method
Fix: Document the VM reservations first. If they are similar, the slot policy is precise; if they differ widely (as across vCenter, NSX, SDDC Manager and VCF Operations), use Cluster resource Percentage or cap the slot with das.slotmeminmb.
Forgetting That Policy Choice Affects Admissibility
Symptom: Recommending a policy without recalculating how many VMs it actually allows
Cause: Separating the 'which policy' decision from the 'how many VMs' consequence
Fix: Always end policy recommendations with concrete capacity statements: 'This slot policy admits 8 VMs with a 32.4 GB slot' or 'This 25% policy admits reservations up to 108,000 MHz and 264 GB'.

Task 3 Design HA Policy for VCF Management Domain

Produces a professional design decision document (DD format) that specifies HA policy, justifies cluster sizing, documents reservation requirements, and addresses what-if failure scenarios.

Create a comprehensive HA design document for the VCF management domain that demonstrates N+1 availability architecture and applies admission control concepts to production VCF components.

Step 1

Identify the management VMs and their resource requirements in the VCF management domain. The inventory below is illustrative and partial (it omits NSX Edge nodes, VCF Automation and Operations for Logs) — take the real figures from the VCF 9.0 Planning and Preparation Workbook and each product's deployment sizing. List: vCenter (Medium: 8 vCPU, 30 GB — vSphere 9.0 docs: Hardware Requirements for the vCenter Appliance), NSX Manager cluster nodes (Medium form factor: 6 vCPU, 24 GB each, 3 nodes = 18 vCPU/72 GB — VCF 9.0.1 docs p.2757), SDDC Manager (4 vCPU, 16 GB — the VCF Installer appliance specification; deployed in the management domain it switches to SDDC Manager mode, VCF 9.0.1 docs p.1273, p.1265), VCF Operations analytics cluster in the High Availability model (3 nodes, p.859; Medium nodes 8 vCPU, 32 GB each = 24 vCPU/96 GB — KB 397782 VCF Operations 9.0 Sizing Guidelines), VCF Operations fleet management appliance (4 vCPU, 12 GB — p.1353) and one collector (Classic Small cloud proxy: 2 vCPU, 8 GB — p.2965 Table 934). Total: 8+18+4+24+4+2 = 60 vCPU and 30+72+16+96+12+8 = 234 GB RAM across the listed management VMs.

Expected

Documented inventory of management domain VMs with resource specifications (figures from the cited sizing sources; confirm against the Planning and Preparation Workbook for your release). Example table:

VMvCPURAM GBNotes
vCenter (Medium)830HA restart; vCenter HA not supported in VCF
NSX Manager-1 (Medium)624Part of 3-node cluster
NSX Manager-2 (Medium)624Part of 3-node cluster
NSX Manager-3 (Medium)624Part of 3-node cluster
SDDC Manager416Single appliance, HA restart
VCF Operations analytics node ×38 each32 eachHigh Availability model (p.859)
VCF Operations fleet management412Single appliance (p.1353)
VCF Operations collector (Classic Small)28Cloud proxy (Table 934)
Total60234
Underestimating management VM resource requirements can cause uncontrolled restarts during host failure when there's insufficient spare capacity.
The three-node NSX Manager cluster needs a majority (2 of 3 nodes) to stay available: it tolerates one node failure, not two. Keep the nodes on separate hosts with a DRS VM-VM anti-affinity rule (VCF-NSX-LM-RCMD-CFG-002), which needs at least four hosts to still hold after a host failure.
Step 2

Calculate the minimum cluster size to maintain N+1 availability for the listed management VMs, using the usable capacity from Task 1 (88 GB and 16 cores per host after the hypervisor's own use). Memory: 234 GB / 88 GB = 2.66 host equivalents; add the N+1 reserve: 2.66 + 1 = 3.66, round up to 4 hosts. vCPU at the 1:1 assumption: 60 vCPU / 16 cores = 3.75 host equivalents; 3.75 + 1 = 4.75, round up to 5 hosts. The limiting resource is vCPU, so the minimum cluster size is 5 hosts.

Minimum cluster sizing calculation: Management workload = 234 GB RAM (2.66 host equivalents at 88 GB usable); with N+1 reserve = 3.66 → 4 hosts. vCPU constraint = 60 vCPU (3.75 hosts at 16 cores); with N+1 = 4.75 → 5 hosts. Limiting factor: vCPU. Minimum cluster size = 5 hosts.
Calculating minimum size for only one resource type (e.g., memory) and ignoring CPU can lead to CPU oversubscription during failover.
Always calculate for both memory and vCPU; use the larger result. Often vCPU is the limiting factor for compute-intensive workloads like NSX Managers.
Step 3

Document the HA design decision (DD format). Create a formal design decision document with sections: (1) Requirement: 'Maintain N+1 availability for VCF management domain to support 99.95% uptime SLA (max 22 minutes downtime/month)'; (2) Constraints: 'Cluster must survive single host failure without VM restart cascades'; (3) Assumptions: 'VMs are distributed across all 5 hosts (DRS anti-affinity for the NSX Manager and VCF Operations nodes)'; (4) Design Decision: 'Deploy a 5-host management cluster with HA admission control for one host failure using percentage-based failover capacity (VCF-CLS-RCMD-CFG-001, VCF 9.0.1 docs p.550); vSphere calculates the reserve from the host failures to tolerate (20% in 5 hosts). Slot policy is rejected because the management VMs carry differing reservations (a 32 GB VCF Operations node sets an oversized slot). Admit only management VMs to this cluster (no user workloads)'; (5) Justification: 'Sizing calculation: 234 GB management workload must fit on the 4 surviving hosts: 4 × 88 GB = 352 GB usable, leaving 118 GB for growth; 60 vCPU on 4 × 16 = 64 cores at 1:1. N+1 compliant on both resources.'

Expected

Formal HA Design Decision document:
VCF Management Domain HA Design (v9.0)
Requirement: N+1 availability for VCF management domain
Current State: 4-host cluster (insufficient for management workload)
Proposed Design: 5-host cluster, HA admission control for 1 host failure, percentage-based (VCF-CLS-RCMD-CFG-001)
Rationale: Management workload totals 234 GB RAM / 60 vCPU; after one host failure the 4 surviving hosts offer 352 GB usable and 64 cores, so the workload fits with 118 GB headroom. Provides N+1 failover coverage on both memory and CPU.
Impact: Requires upgrade from 4-host to 5-host cluster.

Vague designs like 'Add more hosts' without sizing calculations fail VCDX review. Always provide concrete numbers and justifications.
Design decision documents are expected in VCDX defense. Use formal structure: Requirement, Constraint, Assumption, Design Decision, Justification, Risk/Mitigation. Panelists probe how you arrived at your sizing and whether you considered failure scenarios.
Step 4

Test the design with a what-if scenario: what if 2 hosts fail simultaneously (N+2)? The design assumes N+1. With 2 host failures: 5 − 2 = 3 hosts remain. Memory: 3 × 88 GB = 264 GB usable against 234 GB of management VMs — fits. vCPU at the 1:1 assumption: 3 × 16 = 48 cores against 60 vCPU — does NOT fit. Conclusion: the design is N+1 only; it does not survive two host failures without CPU overcommit on the management stack. Document: 'Current HA design supports N+1 (single host failure). N+2 is not a requirement per the business SLA; if an N+2 requirement emerges the cluster grows to 6 hosts (4 survivors × 16 = 64 cores ≥ 60 vCPU, 352 GB ≥ 234 GB).'

What-if analysis: N+2 scenario. With 2 host failures: 3 remaining hosts = 264 GB usable and 48 cores. Management workload = 234 GB / 60 vCPU. Conclusion: memory would survive but vCPU would not at the 1:1 assumption, so the design is N+1 only. Document as 'N+1 validated; N+2 not supported without CPU overcommit'. If the customer requires N+2, update the design to 6 hosts.
Do not claim N+2 support unless both memory and CPU survive the loss of two hosts. Represent the design as N+1 compliant.
VCDX panelists appreciate proactive what-if analysis — but only when the arithmetic covers every constrained resource. Checking memory alone and declaring N+2 'incidentally supported' is exactly the kind of gap a panel probes.

Validation Gate

Check: Verify HA design document and what-if scenario analysis

Expected: Formal design decision document with justification and what-if scenario analysis

Common Errors

Sizing Cluster for Current Workload Only, Ignoring Reserve
Symptom: Calculating minimum cluster size = 3 hosts (enough for 234 GB of management VMs) without adding the N+1 reserve, or sizing on memory only
Cause: Not including the (n-1) calculation; forgetting that 1 host must be reserved
Fix: Always add 1 full host to your sizing: If the workload fits in 2.66 host equivalents of memory you need 3.66 → 4 hosts; repeat for vCPU (3.75 + 1 = 4.75 → 5) and take the larger. Use the formula: workload host equivalents + 1 = minimum sizing, for each resource.
Ignoring NSX Manager Cluster Quorum Requirements
Symptom: Proposing a 3-host management cluster: the 3-node NSX Manager cluster tolerates only one node failure (majority of 2 needed), and the VM-VM anti-affinity rule for the three managers needs at least four hosts to survive a host failure
Cause: Not understanding that NSX Manager needs 3 nodes for HA and the cluster itself needs reserve capacity separate from NSX HA
Fix: NSX Manager HA (3-node cluster) is separate from vSphere cluster HA (N+1 failover reserve). Size the vSphere cluster to hold 3 NSX Managers PLUS reserve 1 host. Total: 3 NSX nodes + other management VMs + 1 reserve host.
Designing Without Documenting Assumptions
Symptom: Providing a cluster size (5 hosts) without explaining assumptions about VM distribution, workload consolidation, or resource allocation
Cause: Jumping to conclusion without showing decision logic
Fix: Always document assumptions explicitly: 'Assume all management VMs distributed across 5 hosts, no single-host concentration.' 'Assume no over-provisioning of vCPU (1:1 allocation).' This allows panelists to probe your reasoning and ask 'what if we over-provision?'

Final Validation

HA admission control lab completed with slot-based vs percentage-based analysis, scenario evaluation, and VCF management cluster design

✓ Admission control calculations (Task 1, Steps 1-4) → Documented slot sizes for all VM profiles, N+1 capacity calculations

✓ Scenario evaluation and policy recommendation (Task 2, Steps 1-4) → 96GB VM admissibility determination, failover impact analysis, slot vs. percentage comparison, and justified policy choice

✓ VCF management design decision (Task 3, Steps 1-4) → Management VM inventory, minimum cluster sizing, formal DD document, N+2 what-if scenario

Cleanup / Restore

• Revert to initial snapshot if any lab VMs were created

• Document all design decisions and calculations for VCDX study

Design Reflection (VCDX)

During a VCDX defense, panelists will probe your HA design with questions like: 'You sized the cluster for 5 hosts. What happens if a host fails during a maintenance window when a 3rd host is offline?' Or: 'How would you handle a scenario where NSX Managers are concentrated on 2 hosts and one host fails?' Be prepared to explain admission control trade-offs, justify why you chose percentage-based over slot-based (VCF-CLS-RCMD-CFG-001), and articulate what would change if requirements shifted to N+2 or if workload profiles expanded. Panelists expect designers to show flexibility—how would you adjust the design if vCF Ops expanded from 64GB to 128GB?

Requirements

  • VCF management domain must maintain N+1 availability for 99.95% uptime SLA
  • Cluster must support heterogeneous management VMs (vCenter, NSX Managers, SDDC Manager, VCF Ops)
  • Admission control must prevent oversized VM admissions that would violate HA during failover
  • Design must be documented in formal DD format with sizing calculations and assumptions
  • Must address minimum cluster size, slot calculation methodology, and what-if failure scenarios

Constraints

  • Each host in the management cluster has fixed capacity (96GB RAM, 16 vCPU cores)
  • NSX Manager requires 3-node cluster for quorum and HA
  • VCF management domain must not share hosts with user workloads (isolation requirement)
  • Admission control policy chosen must apply consistently across all admitted VMs
  • One full host capacity must always be reserved for failover (N+1 requirement)

Assumptions

  • All management VMs will be distributed across available hosts (no single-host concentration)
  • Hypervisor reserves ~8GB per host, leaving 88GB available for VMs
  • vCPU allocation is 1:1 (no oversubscription) for deterministic performance
  • Management VM resource requirements are stable (no dynamic scaling)
  • Slot size is calculated once and remains stable (no mid-cluster addition of different-sized VMs)
  • Network and storage are not the limiting factors (CPU and memory are the constraints)
  • Single host failure is the design-basis failure scenario; N+2 is not required

Risks

  • Risk: Cluster growth adds larger VMs without recalculating slot size. Mitigation: Document when slot size becomes invalid and requires cluster redesign.
  • Risk: VCF Ops or another management VM is unexpectedly resized upward. Mitigation: Establish change control requiring HA impact analysis before VM resize in management domain.
  • Risk: Operational team over-admits VMs by disabling HA or ignoring admission control. Mitigation: Use vSphere admission control policies that are enforced and non-bypassable without explicit approval.
  • Risk: Network or storage becomes the limiting factor instead of CPU/memory. Mitigation: Extend this analysis to validate network (NSX bandwidth) and storage (IOPS, capacity) separately.
  • Risk: A third host fails within 15 minutes of the first host failure (before first host VMs have fully restarted). Mitigation: Monitor first host failure recovery time; if recovery takes >15 min and concurrent failures are possible, increase cluster size or add recovery time SLA.

Self-Assessment Discussion Prompts

  1. You proposed a 5-host management cluster to maintain N+1 availability. A business stakeholder asks: 'Why do we need 5 hosts? We currently run on 4, and nothing has failed in 2 years.' How would you justify the upgrade? What specific risks does the 4-host cluster face during a single host failure?
  2. Compare slot-based vs. percentage-based admission control. In a cluster with heterogeneous VM sizes (2GB to 64GB), which method provides better capacity utilization and why? What are the trade-offs?
  3. VCF Ops is a 64GB VM (the largest in your cluster). A second ops tool is proposed that also requires 64GB. How would your HA design need to change? What's the new minimum cluster size?
  4. Your cluster is sized for N+1 but a compliance requirement emerges: 'Infrastructure must survive simultaneous failure of any 2 hosts.' How do you redesign the cluster? What's the new minimum size and cost implications?
  5. During a post-mortem after a host failure, you discover that all NSX Manager VMs restarted on a single surviving host, overloading it momentarily. Which DRS VM-VM anti-affinity rule (VCF-NSX-LM-RCMD-CFG-002) would have kept the managers apart, and why can no HA admission control policy do that job? Should you separate management VMs across multiple failure zones?
  6. You're asked to design HA for a smaller regional VCF instance with only vCenter and SDDC Manager (no NSX, no Ops). How does this change your admission control calculations? Could a 3-host cluster suffice?

Extensions

Extend to Multi-Region HA Design

Design HA for a VCF environment spanning 2 geographic regions with asynchronous replication. How would you adapt admission control policies? What is minimum cluster size per region?

Integrate Storage Admission Control

Extend the CPU/memory analysis to include storage constraints. If each management VM requires 500GB storage and the cluster has 10TB total, does storage change the minimum cluster size calculation?

Dynamic Workload Scenario

Model a scenario where VCF Ops is expanded from 64GB to 128GB. Recalculate cluster sizing. At what point does slot size change? What remediation steps would you take during the upgrade?

⚠ Known Pitfalls

Confusing admission control reserve with spare capacity. Reserve capacity (N+1) is not spare; it's guaranteed to be held unused in steady state to support failover.
Problem: Teams provisioning 'extra' hosts for growth while failing to reserve 1 host for HA end up with oversubscribed clusters that fail under single host failure.
Resolution: Establish cluster sizing policy: 'Total capacity = Workload + 1 host reserve.' Growth requires adding hosts in pairs or clusters of 3+ to maintain reserve.
Assuming all VMs fit in one slot and using a single cluster-wide slot size. If VMs vary from 2GB to 64GB, using the 64GB slot size wastes capacity for small VMs.
Problem: Capacity is underutilized. A cluster that could admit 10 small VMs may be designed to admit only 4 large VMs, wasting 60% of available capacity.
Resolution: Use different admission control thresholds for different VM classes, or use percentage-based admission control for heterogeneous workloads.
Not communicating HA design decisions to the operations team. If ops doesn't understand why the cluster is sized for N+1, they may disable HA or over-admit VMs to 'get more capacity.'
Problem: Operational teams accidentally reduce HA compliance, violating SLAs unintentionally.
Resolution: Provide training and documentation. Create alerts if admission control is disabled or if admitted VM count exceeds design threshold.
Designing HA for current workload only, without growth buffer. If management workload grows 20% in year 2, the cluster is suddenly oversubscribed.
Problem: Cluster becomes non-compliant with N+1 SLA mid-year, requiring emergency host addition.
Resolution: Size clusters with 20-30% growth headroom: 'Design for 176GB current workload + 50GB forecast growth = 226GB design basis.' Establish annual review cadence to reassess sizing.

References

Was this page useful?
Type to search. ↑ ↓ to move, Enter to open, Esc to close.