Planning and Designing
Design inputs and constraints
Start with requirements: RPO/RTO, availability target, growth, compliance, and budget. Convert them into host count, CPU and memory reservations, storage capacity and IOPS, and network bandwidth. Use the VMware Compatibility Guide (HCL) before selecting servers, adapters, and storage arrays; unsupported components are a common distractor. Size vCenter by inventory scale (hosts and VMs), not by cluster count, and remember the appliance deployment sizes from Tiny to Large. Use Enhanced Linked Mode when multiple vCenter instances must share a single inventory view.
Cluster and availability design
Design clusters around failure domains. N+1 means the cluster survives one host failure; N+2 survives two. Admission control must reserve enough capacity for the chosen policy. Keep management and workload clusters separate so a workload outage cannot consume the management plane. For vSAN, plan disk groups, cache tier, and failure tolerance (FTT=1 needs at least three hosts; FTT=2 needs at least five). For networking, plan VLANs, MTU (jumbo frames end-to-end for vSAN/iSCSI), uplink redundancy, and a separate management network.
Traps
Do not treat a CPU overcommit ratio as available HA capacity. Do not size to the maximums-know them, then leave headroom. Do not assume vCenter HA replaces vSphere HA; vCenter HA protects the management appliance, vSphere HA protects workloads. Do not design DR without testing the runbook.
Worked example: Five hosts, each 512 GB RAM, target 80% utilization. Workload = 5 × 512 × 0.80 = 2,048 GB. If one host fails, four hosts provide 2,048 GB physical-zero headroom, so N+1 is not truly satisfied at 80%. Lower the target to 75% (1,920 GB) or add a sixth host.
Rule of thumb: Design to the compatibility guide, size to N+1 or N+2, and verify that surviving hosts can carry the reservations.