Data Protection and Disaster Recovery
Protection primitives
A Protection Domain (PD) groups VMs so they are snapshotted consistently together; schedule local snapshots and optionally replicate them to a remote cluster. In Prism Central, the equivalent higher-level construct is a protection policy applied through categories.
Replication options, from least to most strict:
- Async DR - periodic replication; RPO measured in minutes. Guest data is crash-consistent unless NGT is installed for VSS quiescing.
- NearSync - continuous-ish replication with an RPO of a few minutes (typically 1–5), still asynchronous.
- Metro Availability - synchronous replication across a stretched cluster with a witness; RPO is effectively zero; both sites serve reads/writes while healthy.
Nutanix Leap orchestrates DR: runbooks, one-click failover, and non-disruptive testing of the DR plan. For long-term retention beyond Nutanix snapshots, third-party backup products or object storage targets are used.
Traps
- Mixing up RPO (how much data you can afford to lose) with RTO (how long you can be down). Questions often hinge on which one is being described.
- Claiming Metro gives RPO zero without a witness - the witness is what prevents split-brain.
- Assuming PD snapshots are application-consistent by default. They are not, without NGT.
- Forgetting that replication targets need capacity, network bandwidth, and matching network configuration.
Worked example
A database team requires zero data loss and automatic continuation if one site fails → Metro Availability with a witness, not Async DR. If they can tolerate a few minutes of loss but need fast recovery, NearSync plus Leap runbooks is the fit.