Synthetic Backups

---

Real-Time Backup

Real-time backup is another feature that combines different technologies. Here the data is backed up (or replicated) as it changes, resulting in continuous backups. The data is replicated to a second host and then backed up from the second (target) host to tape. This moves the backup burden to the target system in exchange for a steady impact on the source system and the network. Real-time backup also offers you the opportunity to create versions based on more system-based events. Markers would be used in the log stream to synchronize the source and target checkpoints with application quiescence. Some solutions today actually sit at the application layer, above the cache, and will replicate changes incrementally, rather than the entire filesystem. These software solutions may provide a higher level of protection in the event of a failure, since the replication may be committed before the actual source information is committed.

Backup and Replication

Backup of a filesystem is typically accomplished by copying files from the source hard disk across the network to a network server hosting tape drives. The backup system would need to 'walk' the filesystem in order to accomplish its task. The frequency of the backups is governed by practical considerations such as network impact, system impact, and available media. On the other hand, remote filesystem replication products monitor changes to file objects and copy these changes to a remote disk location for safekeeping. Some solutions choose to do a full file copy, block-level changes, or simply incremental or delta changes of a file when they sense a file change. The remote replica can be utilized for disaster recovery in the case of major physical damage to the primary filesystem. Data protection in this scenario occurs in real time as objects change. Its use is usually governed by network impact, available replica storage, and overhead on the server. Figure 10.2 is an example of a real-time backup configuration.

Figure 10.2: Real-time backup-High level.

What if we replicate only the block changes? Well, that's being done today, with VERITAS Volume Replicator, FalconStor's IPStor, EMC, Hitachi, and other solutions. What if we were able to back up only those changes for an infinite recovery level based on the retention of those blocks written to tape, as shown in Figure 10.3? Taking the block-level replication to the next level may include an option to integrate with the backup software that allows only those incremental changes to be written to tape, either continuously or scheduled. In the future, we may have the ability to recover a particular file from a particular point in time from blocks that are backed up or archived to tape after being migrated from the disk-based solution, providing an infinite level of recovery. The challenges that we will continue to face will be the backup of these very large databases and filesystems, and how to best protect this critical data while ensuring we have a more than reasonable recovery time in the event of online data loss. The trade-off with this type of solution is the dependence on the potentially proprietary nature of the data written to tape depending on the backup software with which this solution was integrated. However, in this solution, disk would be the primary backup medium and the tape backup would only be for the backup server's purposes.

Figure 10.3: Future real-time backup-Block-level replication.

Replication will play a major role in the future of backup, especially as we continue to see the data growth scale higher and higher. Many of these replication solutions have been around for quite some time, but because of our general business requirements, we either didn't require that level of availability, found it to be cost-prohibitive, or simply didn't know it existed and worked as well as some of the proprietary hardware solutions. One of the products that has found success in the mid-tier level is NSI Software's Double-Take, shown in Figure 10.4. It will replicate a set of data at the application level, before it reaches the cache. It has been around for quite some time, but has not gotten proper attention because general business requirements have not been sufficiently compelling to evaluate this as a solution. Now we are facing new challenges and new requirements that ask us to stage data in two or three different locations: source, electronic vault, and perhaps a third location for sanity purposes. Solutions like Double-Take allow you to stage the data in such a manner, allowing for more granularity in the replicated data sets. For instance, the first set of data to the staging server might only replicate a subset of that data to the next stage, and so on.

Figure 10.4: Double-Take replication.

It is assumed that clients will target a remote protection server that will host replicas of their data. These replicas will be backed up to tape via normal scheduled backups. In this scenario, the protection server acts as a disk consolidation server and then performs optimized backups from this consolidated storage. This type of backup does not require a SAN, but the protection server storage certainly could justify SAN connectivity. The replica data on the protection server can be backed up using traditional backup methods or any of the supported frozen image backups.

An anticipated way to further leverage technologies and get even more benefits from real-time backup is to use injected markers in the replication stream to notify the remote target or protection server that the application has been quiesced and that it is safe to take a snapshot. This allows the creation of consistent point-in-time snapshots that can potentially be used for rollback and can be written to tape. Figure 10.5 shows an example of this.

Figure 10.5: Consistent snapshots.

Many different scenarios based on the combination of these technologies are being investigated, and we are sure we will see many of them come to market in the not too distant future. The ability to create consistent versions of data to protect against corruption while allowing for disk-based rollback to a point in time is very attractive. Coupled with this is the ability to move data to tape for true data protection with minimal impact on production systems and applications-a very attractive feature.

---