Incorporated into the Prism management console AOS presents synchronous and asynchronous options to protect virtual machines (VMs). The process uses re-direct on write (ROW) snapshots at the Nutanix Distributed Storage Fabric (NDFS) level and doesn’t require any interaction with the host’s hypervisor. So why isn’t the hypervisor used, they have snapshot APIs? Hypervisor snapshots use copy on write (COW), these are managed by the hypervisor and are typically more disk intensive.
The following sections break down the different terminology and actions.
Protection operates at a per-VM and/or group of VMs level opposed to capturing an entire storage pool – think LUN level. This way VMs can be logically collected, have custom / tailored snapshot and replication schedules applied. Nutanix refers to this as Protection Domains. It is worth noting too that Protection Domains don’t need to have their snapshots replicated anywhere, they can reside on the local Nutanix cluster for point in time rollbacks either in-place or to alternate location on the same cluster.
Nutanix doesn’t require hardware accelerators to move data between data centres which means the replication options are defined by the WAN bandwidth and available cluster capacity at the ‘opposite’ site(s). Supported topologies:
- One to One (uni-directional, bi-directional & Cloud Connect)
- One to Many
- Many to One
Operational characteristics of the WAN can be respected with bandwidth optimisation (think throttling) and operational windows defined, data is SSL wrapped and compressed before transmission.
Asynchronous and synchronous is supported in varying levels for each hypervisor, the table below lists each hypervisor and their respective supported functionality.
A couple of points to note: Asynchronous replication can be scheduled to a frequency down to 1 minute as of the AOS 5.1 release, this feature is also referred to as Light Weight Snapshots. Synchronous replication requires a round trip time (RTT) of less than 5 milliseconds (ms).
(Nutanix has alluded to the availability of synchronous replication for AHV during Q2CY2018)
Public cloud consumption isn’t acceptable for all organisations however, not all organisations have more than one location to send their VMs / data too. When defining a ‘Remote Site’ in Prism credentials can be supplied to connect to an Amazon Web Service (AWS) S3 storage instance or Microsoft’s Azure Storage Blob. Part of this service also requires a VM to reside in the compute instance of the cloud provider, this is for a Nutanix Controller VM to act as the receiver for data ingress.