One commonly asked question about the use of snapshots as a data protection strategy is the fact that the "real" data and the snapshot data are located on the same drives. Loss of those drives would result in the loss of both the primary data and the backup.

This is a valid concern. Local snapshots are used for day-to-day backup and recovery needs, and in that respect the snapshot is a backup. Close to 99% of all recovery scenarios in NetApp environments rely on snapshots to meet even the most aggressive RTO requirements.

Local snapshots should, however, never be the only backup strategy, which is why NetApp offers technology such as SnapMirror and SnapVault replication to quickly and efficiently replicate snapshots to an independent set of drives. In a properly architected solution with snapshots plus snapshot replication, the use of tape can be minimized to perhaps a quarterly archive or eliminated entirely.

There are many options for using ONTAP Snapshot copies to protect your data, and snapshots are the basis for many other ONTAP features, including replication, disaster recovery, and cloning. A complete description of snapshot technology is beyond the scope of this document, but the following sections provide a general overview.

There are two primary approaches to creating a snapshot of a dataset:

A crash-consistent backup of a dataset refers to the capture of the entire dataset structure at a single point in time. If the dataset is stored in a single volume, then the process is simple; a Snapshot can be created at any time. If a dataset spans volumes, a consistency group (CG) snapshot must be created. Several options exist for creating CG snapshots, including NetApp SnapCenter software, native ONTAP consistency group features, and user-maintained scripts.

Crash-consistent backups are primarily used when point-of-the-backup recovery is sufficient. When more granular recover is required, application-consistent backups are usually required.

The word "consistent" in "application-consistent" is often a misnomer. For example, placing an Oracle database in backup mode is referred to as an application-consistent backup, but the data is not made consistent or quiesced in any way. The data continue to change throughout the backup. In contrast, most MySQL and Microsoft SQL Server backups do indeed quiesce the data before executing the backup. VMware may or may not make certain files consistent.

The term "consistency group" refers to the ability of a storage array to manage multiple storage resources as a single image. For example, a database might consist of 10 LUNs. The array must be able to back up, restore, and replicate those 10 LUNs in a consistent manner. Restoration is not possible if the images of the LUNs were not consistent at the point of backup. Replicating those 10 LUNs requires that all the replicas are perfectly synchronized with each other.

The term "consistency group" is not often used when discussing ONTAP because consistency has always been a basic function of the volume and aggregate architecture within ONTAP. Many other storage arrays manage LUNs or file systems as individual units. They could then be optionally configured as a "consistency group" for purposes of data protection, but this is an extra step in the configuration.

ONTAP has always been able to capture consistent local and replicated images of data. Although the various volumes on an ONTAP system are not usually formally described as a consistency group, that is what they are. A snapshot of that volume is a consistency group image, restoration for that snapshot is a consistency group restoration, and both SnapMirror and SnapVault offer consistency group replication.

Consistency group snapshots (cg-snapshots) are an extension of the basic ONTAP Snapshot technology. A standard snapshot operation creates a consistent image of all data within a single volume, but sometimes it is necessary to create a consistent set of snapshots across multiple volumes and even across multiple storage systems. The result is a set of snapshots that can be used in the same way as a snapshot of just one individual volume. They can be used for local data recovery, replicated for disaster recovery purposes, or cloned as a single consistent unit.

The largest known use of cg-snapshots is for a database environment of approximately 1PB in size spanning 12 controllers. The cg-snapshots created on this system have been used for backup, recovery and cloning.

Most of the time, when a data set spans volumes and write order must be preserved, a cg-snapshot is automatically used by the chosen management software. There is no need to understand the technical details of cg-snapshots in such cases. However, there are situations in which complicated data protection requirements require detailed control over the data protection and replication process. Automation workflows or the use of custom scripts to call the cg-snapshot APIs are some of options. Understanding the best option and the role of cg-snapshot requires a more detailed explanation of the technology.

Creation of a set of cg-snapshots is a two-step process:

Write fencing is established serially. This means that as the fencing process is set up across multiple volumes, write I/O is frozen on the first volume in the sequence as it continues to be committed to volumes that appear later. This might initially appear to violate the requirement for write order to be preserved, but that only applies to I/O that is issued asynchronously on the host and does not depend on any other writes.

For example, a database might issue a lot of asynchronous datafile updates and allow the OS to reorder the I/O and complete them according to its own scheduler configuration. The order of this type of I/O cannot be guaranteed because the application and operating system have already released the requirement to preserve write order.

As a counter example, most database logging activity is synchronous. The database does not proceed with further log writes until the I/O is acknowledged, and the order of those writes must be preserved. If a log I/O arrives on a fenced volume, it is not acknowledged and the application blocks on further writes. Likewise, file system metadata I/O is usually synchronous. For example, a file deletion operation must not be lost. If an operating system with an xfs file system deleted a file and the I/O that updated the xfs file system metadata to remove the reference to that file landed on a fenced volume, then the file system activity would pause. This guarantees the integrity of the file system during cg-snapshot operations.

After write fencing is set up across the target volumes, they are ready for snapshot creation. The snapshots need not be created at precisely the same time because the state of the volumes is frozen from a dependent write point of view. To guard against a flaw in the application creating the cg-snapshots, the initial write fencing includes a configurable timeout in which ONTAP automatically releases the fencing and resumes write processing after a defined number of seconds. If all the snapshots are created before the timeout period lapses, then the resulting set of snapshots are a valid consistency group.

There are two primary approaches to snapshot-based backups:

A crash-consistent backup of a database refers to the capture of the entire database structure, including datafiles, redo logs, and control files, at a single point in time. If the database is stored in a single volume, then the process is simple; a Snapshot can be created at any time. If a database spans volumes, a consistency group (CG) snapshot must be created. Several options exist for creating CG snapshots, including NetApp SnapCenter software, native ONTAP consistency group features, and user-maintained scripts.

Crash-consistent Snapshot backups are primarily used when point-of-the-backup recovery is sufficient. Archive logs can be applied under some circumstances, but when more granular point-in-time recovery is required, a online backup is preferable.

The basic procedure for a snapshot-based online backup is as follows:

This procedure yields a set of snapshots containing datafiles in backup mode and the critical archive logs generated while in backup mode. These are the two requirements for recovering a database. Files such as control files should also be protected for convenience, but the only absolute requirement is protection for datafiles and archive logs.

Although different customers might have very different strategies, almost all of these strategies are ultimately based on the the same principles outlined below.

When designing volume layouts for Oracle databases, the first decision is whether to use volume-based NetApp SnapRestore (VBSR) technology.

Volume-based SnapRestore allows a volume to be almost instantly reverted to an earlier point in time. Because all of the data on the volume is reverted, VBSR might not be appropriate for all use cases. For example, if an entire database, including datafiles, redo logs, and archive logs, is stored on a single volume and this volume is restored with VBSR, then data is lost because the newer archive log and redo data are discarded.

VBSR is not required for restore. Many databases can be restored by using file-based single-file SnapRestore (SFSR) or by simply copying files from the snapshot back into the active file system.

VBSR is preferred when a database is very large or when it must be recovered as quickly as possible, and the use of VBSR requires isolation of the datafiles. In an NFS environment, the datafiles of a given database must be stored in dedicated volumes that are uncontaminated by any other type of file. In a SAN environment, datafiles must be stored in dedicated LUNs on dedicated volumes. If a volume manager is used (including Oracle Automatic Storage Management [ASM]), the diskgroup must also be dedicated to datafiles.

Isolating datafiles in this manner allows them to be reverted to an earlier state without damaging other file systems.

For each volume with Oracle data in a SAN environment, the percent-snapshot-space should be set to zero because reserving space for a snapshot in a LUN environment is not useful. If the fractional reserve is set to 100, a snapshot of a volume with LUNs requires enough free space in the volume, excluding the snapshot reserve, to absorb 100% turnover of all of the data. If the fractional reserve is set to a lower value, then a correspondingly smaller amount of free space is required, but it always excludes the snapshot reserve. This means that the snapshot reserve space in a LUN environment is wasted.

In an NFS environment, there are two options:

With the first option, percent-snapshot-space is set to a nonzero value, typically around 20%. This space is then hidden from the user. This value does not, however, create a limit on utilization. If a database with a 20% reservation experiences 30% turnover, the snapshot space can grow beyond the bounds of the 20% reserve and occupy unreserved space.

The main benefit of setting a reserve to a value such as 20% is to verify that some space is always available for snapshots. For example, a 1TB volume with a 20% reserve would only permit a database administrator (DBA) to store 800GB of data. This configuration guarantees at least 200GB of space for snapshot consumption.

When percent-snapshot-space is set to zero, all space in the volume is available to the end user, which delivers better visibility. A DBA must understand that, if he or she sees a 1TB volume that leverages snapshots, this 1TB of space is shared between active data and Snapshot turnover.

There is no clear preference between option one and option two among end users.

Oracle Doc ID 604683.1 explains the requirements for third-party snapshot support and the multiple options available for backup and restore operations.

The third-party vendor must guarantee that the company's snapshots conform to the following requirements:

ONTAP and NetApp Oracle management products comply with these requirements.