The basic unit of high availability is the HA pair. Each pair contains redundant links to support replication of data to NVRAM. NVRAM is not a write cache. The RAM inside the controller serves as the write cache. The purpose of NVRAM is to temporarily journal data as a safeguard against unexpected system failure. In this respect, it is similar to a database redo log.

Both NVRAM and a database redo log are used to store data quickly, allowing changes to data to be committed as quickly as possible. The update to the persistent data on drives (or datafiles) does not take place until later during a process called a checkpoint on both ONTAP and most databases platforms. Neither NVRAM data nor database redo logs are read during normal operations.

If a controller fails abruptly, there are likely to be pending changes stored in NVRAM that have not yet been written to the drives. The partner controller detects the failure, take control of the drives, and applies the required changes that have been stored in NVRAM.

Takeover and giveback refers to the process of transferring responsibility for storage resources between nodes in an HA pair. There are two aspects to takeover and giveback:

Network interfaces supporting CIFS and NFS traffic are configured with both a home and failover location. A takeover includes moving the network interfaces to their temporary home on a physical interface located on the same subnet(s) as the original location. A giveback includes moving the network interfaces back to their original locations. The exact behavior can be tuned as required.

Network interfaces supporting SAN block protocols such as iSCSI and FC are not relocated during takeover and giveback. Instead, LUNs should be provisioned with paths that includes a complete HA pair which results in a primary path and a secondary path.

The second aspect of takeover and giveback is the transfer of disk ownership. The exact process depends on multiple factors including the reason for the takeover/giveback and the command line options issued. The goal is to perform the operation as efficiently as possible. Although the overall process might appear to require several minutes, the actual moment in which ownership of the drive is transitioned from node to node can generally be measured in seconds.

Host I/O experiences a short pause in I/O during takeover and giveback operations, but there should not be application disruption in a correctly configured environment. The actual transition process in which I/O is delayed is generally measured in seconds, but the host might require additional time to recognize the change in data paths and resubmit I/O operations.

The nature of the disruption depends on the protocol:

The expected takeover times during which the storage system cannot serve data to an application environment are shown in the table below. There should not be any errors in any application environment, the takeover should instead appear as a short pause in IO processing.