Design and Plan a Nutanix Multicloud Infrastructure
This domain asks you to turn a workload requirement - capacity, RPO/RTO, latency, growth - into a node count, resiliency factor, network topology and Prism Central footprint. Expect scenario questions phrased as "minimum number of nodes" or "which design meets the requirement".
Resiliency drives the arithmetic
- RF2 stores two copies of every extent and tolerates one host or disk failure; RF3 stores three and tolerates two.
- Usable capacity is roughly raw divided by the resiliency factor. Budget N+1 so the cluster can rebuild at full performance after losing a node.
- Two-node clusters and stretched (metro) clusters need a witness at a third site; without it the surviving site cannot arbitrate a split brain.
Sizing the control plane
Prism Central runs as one VM (small footprint) or three VMs (scale-out). Size it against VM count and feature usage, give each PC VM a static address plus a VIP, and confirm DNS, NTP and management VLAN reachability before deployment.
Worked example
Requirement: 30 TB usable, survive one host failure, 20% growth room.
- 30 TB × 1.2 = 36 TB usable target.
- RF2 doubles raw: 72 TB raw.
- Rebuild headroom: after losing a node the survivors must still hold 72 TB raw. Four nodes × 24 TB raw = 96 TB; lose one, 72 TB remains - exactly at the limit with no margin. Five nodes × 20 TB = 100 TB raw; lose one, 80 TB remains, leaving about 4 TB usable margin. Answer: five 20 TB nodes, or fewer larger nodes only with a documented margin.
Traps
- Quoting raw capacity as usable, or treating dedup/compression savings as guaranteed headroom.
- Designing a stretched cluster with no witness, or placing the witness inside one of the two sites.
- Selecting balance-tcp bonding without an LACP port-channel on the switch.
- Forgetting that Prism Central consumes cluster resources when hosted on the cluster it manages.