Database Protection Turns a Corner

The seemingly never-ending quest for cost-effective, reliable Oracle data protection and availability has taken a turn for the better. At least two vendors are delivering a new capability that integrates the features of disk-to-disk backup and continuous data protection (CDP) without increasing storage capacity requirements.

This new development can be described as the active target technique. Implementations are currently available in NetApp's Open Systems Snapvault and Syncsort's Backup Express.

Let's take it from the top: The architecture of an active target is similar to that for traditional backup. First, there are IP- or SAN-connected servers directing data to a backup server that has an active target software module installed. Attached to the backup server is a disk array. This disk serves a LUN to the backup server, so it must attach either via iSCSI or Fibre Channel.

For basic recoveries using an active target technique, this configuration is all you need. For some of the advanced features described later in this article, the link between the servers being backed up and the backup server must be a block-level iSCSI or FC SAN connection, not merely a LAN one.

The basic problem addressed by the active target approach is that backup applications, even when backing up to disk, tend to store the data that they are backing up in their own file format, which essentially bundles up a set of files into a larger file. Before we had disk as a viable backup target, this was a logical way to achieve good performance when backing up to tape. However, when backing up to disk it makes less sense, because the data in backup format must go through a recovery process before it can be used by the application.The problem with this recovery process is that it is time-consuming. First, the data has to be read out of the backup format, transported across the network (SAN or IP), and then written to disk.

In a large Oracle database environment, the transfer speed involved in this procedure is an issue, as is the fact that all the data has to be written to a disk, which is typically running RAID 5. All of this is going to slow the recovery effort.

Further, once the data is restored, then all the transaction logs since the last backup have to be replayed. With traditional backups, this can be as far back as the previous night, which is again going to slow down the recovery effort.

This is not the case with backup applications that can create an active target. By leveraging disk as more than just fast tape,” active target-equipped backup apps can create a point-in-time image of the volumes they are backing up.

Not CDP

Many people will confuse active targets with continuous data protection (CDP). There is a difference: Active target software is part of the backup application, while most CDP packages are not.Most CDP solutions are capturing changes as they occur, essentially acting as asynchronous mirrors. This typically raises concerns about the quality of the data being sent to the target. Since the servers being protected are often live databases or email servers, you would prefer to put them in a backup mode to insure a clean snapshot of the data.

Unfortunately, this can’t be done every second. There is often an uneasiness about performance impact as well. Because of these concerns, many customers will inactivate (or at least ignore) the constant protection feature, instead choosing specific points at which to perform the protection task. This allows them to snapshot the server to make sure they have an environment that is “in sync” and to determine when the server is not too busy doing other tasks.

In contrast, because active targets are part of the backup process, they can be set up to back up data far more frequently, such as every hour, or as required, as opposed to once per night or week.

Another shortcoming of CDP is that due to the lack of integration to the backup process, data recovery requires manual interaction with the file system. Because active targets are integrated as part of the overall backup process, traditional recovery remains the same, although faster.

Case in point

Syncsort's Backup Express software, for example, creates a backup image by doing a block-level incremental backup of the volumes it is protecting. This is particularly valuable in an Oracle environment, allowing you to quickly back up just the changed blocks of information multiple times throughout the day, as opposed to backing up the whole database once per night.By using significantly less database server and network bandwidth, the solution provides more recent and more granular data protection points, thereby minimizing data loss and the time spent replaying transaction logs.

The active target is “snapshotted” at the end of each backup to provide versioning, the ability to roll back to a specific incarnation of the database.

For example, if you have a 1-Tbyte-sized Oracle database, the first time you back it up, that entire Tbyte is written to the active target application. The blocks of data that make up that Tbyte are then “snapshotted.” An hour later when you back up that database again, only the changed blocks are sent across the network. They are then written to disk and “snapshotted” again. The subsequent backups can then be performed hourly with a minimal increase in storage capacity. The software can present the database as the original image or as any of the subsequent versions that have been updated on an hourly basis.

Next Page: Rapid access to recovery data Contents:

Rapid access to recovery data

The granularity offered by the active target technique is critical in Oracle environments. Because you are able to do low-resource backups frequently, like every hour, you significantly reduce the transaction playback window that in many cases is the longest part of the overall recovery effort.From a recovery standpoint, because the data is not locked in a backup format, you have rapid access to it. In the event of a volume failure or database corruption, you can perform a “data in-place recovery.” Without an active target, even if you had a recent and viable backup, you would still have to recover data across the network to the Oracle server, which can be a very time-consuming process.

What's more, with an active target in place, you can benefit from functionality that you do not typically get from a backup application. Backup volumes are now smaller, as only the changed blocks are sent across the network and are stored on disk.

With “in-place recovery,” you are able to make one of the above backup versions live and simply point the database at the most recent, viable instance of the database. You are then up and running.

If you have a total server failure, the entire image is stored on disk and rapid recovery of the entire server is possible from a boot CD. Since boot from SAN is used sparingly and most servers in an environment are not even on the SAN, the ability to boot from a CD and perform a full server recovery is critical.

While some of the above capabilities can be achieved with a SAN-based snapshot, SAN snapshots are problematic. The first challenge to SAN-based snapshots is that for them to be effective, all the data, including boot data, must be on the SAN. As mentioned above, this is seldom the case.Secondly, most SAN-based snapshots take an immediate performance impact on the overall array, so, as a result, most storage managers use them in a limited fashion, maybe only a few per day. Lastly, they have no integration when moving to tape. While scripting can be done to develop a very basic level of integration, there are problems in getting everything in sync – and then of course there is the issue of maintaining scripts. These are just a few of the considerations you need to keep in mind.

The active target also does NOT eliminate the need for devices that can do data de-duplication. In fact, they supplement the ability by being able to "push" live copies of the file system in an automated fashion to the de-duplication appliance. This appliance can then store the data for longer retention periods. This combination builds the ideal backup environment: quiet host processing, frugal use of the network backbone, and efficient balance of instant recovery and long-term retention.

Next Page: Leveraging active targets for volume cloning

Leveraging active targets for volume cloning

A backup software application that can create an active target can be leveraged for more than just the backup and recovery process. It can also be used to create so-called thin clones, which are again particularly valuable in an Oracle environment.

A typical Oracle environment needs multiple copies of the same database information. There is, of course, the production data, multiple test/development copies of the data, and the ever-necessary backup copies. All of these extra copies can place quite a burden on the storage administrators and often are not forecasted in the initial design of the storage for a database. Sticking with the 1-Tbyte database above, this can be an additional 3 to 6 Tbytes of copies, but most of it has to be on active storage since the data is still being accessed (especially when it's used for testing and development).In addition to the storage consumption costs, there is the problem with the time it takes to create and re-create these additional data sets. Even when copying from Fibre Channel array to Fibre Channel array, creating multiple instances of a 1-Tbyte environment is time consuming. There are also the soft costs of the administrators, both storage and database, required to manage these copies.

With active targets, you can create a clone from one of the backup images. The snapshot can be presented to the test/dev server and set to track new changes in a read/write mode. This has several benefits.

First, it is an enormous storage consumption cost saving since the core database information is "shared," and then only changes or additions to that data are tracked. This can pull the 6-Tbyte number mentioned above to only a few hundred gigabytes. Secondly, these images can be created almost instantly. No time is wasted waiting for copies to happen or trying to find space somewhere to make the copy.

Lastly, thin clones can also be “rolled forward.” Rolling forward secures the original data so that changes made to it will not harm the original core set of data. In a roll forward situation, for example, if you have a bug in the software and figure out how to fix it in the snapshot, instead of recreating this fix on the production volume you can simply roll this volume forward as the primary volume and be back up and running with the fix in place.

In summary

Active targets move disk-to-disk backup to a higher level of efficiency, while providing for more granular data protection. Backups can happen on the hour instead of once per night. High-quality protection can be more widespread and no longer be limited to just a SAN or NAS. Recoverability is improved, not only from volume corruption, but from total system failure as well.Active targets, as implemented by Syncsort and NetApp, provide the ability to either point to or directly use an instance of a backed up volume. You could also perform a full bare metal recovery in the event of a file server crash. Lastly, the backup process can be amortized across more than just backup and recovery. With cloning, you can use the backup data set for test/development, reporting, auditing and a host of other functions, while at the same time freeing up primary storage resources.

For the first time, the backup and recovery process can save hard costs as well as saving your company. I call that a trend worth watching.

— George Crump, Founder and President, Storage Switzerland