Application consistency
Appliance based replication
Array based replication
Consistency groups
Continuous data protection (CDP)
Continuous remote replication (CRR)
Crash consistency
Deduplication
Five 9s availability
Input/Outputs per second (IOPS)
Point-in-time data recovery
Redundant Array of Independent Nodes (RAIN)
Recovery Point Objective (RPO)
Recovery Time Objective (RTO)
Replication
Snapshot
Write Splitter

Application consistency
Guarantees that the application's data is in a consistent state at the point-in-time when it is replicated, backed-up or snapped. The application will be placed into a quiescent state which will commit in-memory transactions and then halt writes to the database and log files.

For example Microsoft Volume Shadow Copy Service (VSS) provides a framework for application consistent backup, snapshot and replication of Microsoft Exchange and SQL Server. The alternative is crash consistency.

Appliance based replication
Replication is performed by an appliance connected either to the hosts, SAN Fabric or storage array. Two options are available:

Out-of-band
The appliance resides outside of the primary data path therefore an application’s I/O does not flow through the appliance. Implementing an out-of-band appliance delivers replication without impacting an application’s I/O operations (i.e. performance and reliability).

In-band
The appliance resides in the primary data path therefore an application’s I/O flows through the appliance. Implementing an in-band appliance delivers replication which can have an impact on an application’s I/O operations (i.e. performance and reliability).

Array based replication
Replication is performed within the storage processors of the array, no additional hardware or host resources are required.

Consistency groups
A collection of volumes that when replicated remain in a consistent state with respect to each other. All data will be synchronised to the exact same point-in-time. For example you can protect data from one or more applications that use multiple volumes.

Continuous data protection (CDP)
Automatically saves a copy of every change made to a volume locally, essentially capturing every version of the data that the user saves, allowing the administrator to restore data to any point-in-time. Application consistency points can be periodically scheduled to avoid having to recovery from a crash consistent image. The RPO is zero.

CDP is different from traditional backups in that there are no backup schedules and you don't have to specify the point-in-time to which you would like to recover until you are ready to perform a restore. Traditional backups can only restore data to the point at which the backup was taken.

Continuous remote replication (CRR)
Automatically saves a copy of every change made to a volume remotely, essentially capturing every version of the data that the user saves, allowing the administrator to restore data to any point-in-time. Application consistency points can be periodically scheduled to avoid having to recovery from a crash consistent image. The RPO is typically seconds or greater and it supports unlimited distances between storage devices. Two options are available:

Continuous asynchronous
Each write transaction is acknowledged locally at the source side and then sent to the target side. The primary advantage of continuous asynchronous replication is its ability to provide synchronous-like replication without degrading the performance of host applications.

Near continuous snapshots
Transfers data that has changed between one consistent image of the storage subsystem and the next. The use of high-frequency snapshots largely overcomes the shortcomings of the snapshot not being up-to-date. Typically powerful bandwidth reduction compression technologies can be applied resulting in a significant savings in bandwidth.

Crash consistency
The application's data is not put into a consistent state when it is replicated, backed-up or snapped. The data is in the same state as if there had been a power outage, hardware failure or software crash. Most applications have a built-in crash recovery mechanism that will allow it to recover a crash consistent copy of its data. The alternative is application consistency.

Deduplication
Enterprise data is highly redundant, with identical files and sub-file data segments stored within systems. Deduplication solutions assign each data segment a unique ID, based on its content, which is used to compare it with other data segments that have already been backed up. Only new, unique data segments are stored and typically deduplication occurs across sites and servers, hence the term 'global deduplication’.

Deduplication can occur at the data source or the backup target. With source-based deduplication, data is deduplicated as the backup process begins and before the data is sent over the network. This provides the benefit of shorter backup windows and lowered bandwidth requirements, making it ideal for remote or WAN-based backup, VMware, large file servers, and other environments where the backup process is hampered by network or other resource bottlenecks.

For target deduplication the main challenge being addressed is the growth of back-end storage. The backup application sends data to the target storage device and the data is deduplicated at the device, either immediately or at a scheduled time. It is found in VTLs and LAN backup to disk appliances or platforms and provides the benefit of plug and play with existing backup applications. Unlike source based deduplication this will not remove bottlenecks in getting the data to the backup storage device.

Five 9s availability
The equivalent of an average of 5.26 minutes of unplanned downtime per year, or 99.999% system availability. Nowadays this level of enterprise class availability is required for most critical business data.

Input/Outputs per second (IOPS)
The total number of reads (typically around 70%) and writes (typically around 30%) per second provided by a disk system.

Fibre Channel/SAS disks can provide twice the IOPS provided by SATA disks and Flash disks can provide thirty times the IOPS provided by Fibre Channel/SAS disks.

Point-in-time data recovery
Journals all data changes to a dedicated volume allowing recovery to any point-in-time. For example if a volume is corrupt then it could be recovered to the point prior to the corruption occurring. Also see CDP.

Redundant Array of Independent Nodes (RAIN)
RAIN works in a similar fashion to RAID to deliver high availability, but rather than protecting against disk failure it protects against server failure. It uses a grid architecture, which allows for online expansion for increased scalability.

Recovery Point Objective (RPO)
The acceptable amount of data as defined by an organisation that can be lost in the event of a disaster measured in time. For example an RPO of 2 hours requires the data to be restored at a point-in-time no earlier than 2 hours prior to the disaster occurring.

The RPO in conjunction with the RTO is the basis on which a business continuity strategy is developed.

Recovery Time Objective (RTO)
The duration of time as defined by an organisation within which a business process must be restored after a disaster. For example an RTO of two hours requires systems to be back up and running and accessible within 2 hours of the disaster occurring.

Replication
The process of copying or mirroring data from one storage device to another, within the same storage array, or to a different array located locally or remotely. Typically only protects the most recent copy of the data and if it becomes corrupted will simply "protect" the corrupt data. CDP will protect against the effects of data corruption by allowing a restore to a previous, uncorrupted version. Two options are available:

Synchronous
Guarantees zero data loss by mirroring writes to a secondary storage device. A write is not considered complete until acknowledged by both storage devices. Performance drops proportional to distance, as latency increases, therefore it is only suitable when there is limited distance (100km or less) between storage devices. The RPO is zero.

Asynchronous
The write is considered complete as soon as the primary storage device acknowledges it. The secondary storage device is updated, but lags behind the primary. Performance is not impacted therefore it supports unlimited distances between storage devices. The RPO is typically 30 minutes or greater.

Snapshot
A copy of a set of files and directories as they were at a particular point-in-time. Snapshots can be mounted read-only, or read-write, used to instantly restore the current data to a given point-in-time, and can be used for parallel processing such as accelerated backups, reporting and testing. Two options are available:

Logical view
Maintains a log of changes and combines the production volume with these changes to create a logical point-in-time volume. Takes seconds to create and requires significantly less space than a clone.

Clone
Physically independent full copy of the production volume. Can take a considerable amount of time to initially create and requires the same space as the production volume.

Write Splitter
Replicates data to a secondary storage device by intercepting application writes. Options include:

Host operating system
Requires an agent to be installed on each server and therefore will have a small impact on CPU utilisation on the host.

Intelligent Fabric
Provided within the FC switch, from vendors such as Brocade and Cisco, therefore will not have any impact on host performance and will not require the installation of a host agent.

Storage array
Provided within the array's storage processor therefore will not have any impact on host performance and will not require the installation of a host agent.