Reducing the Backup Window With Sun StorEdge Instant Image Software
This article discusses the advantages and methods of using a point-in-time (PIT) type of backup system versus a more traditional backup approach that requires extended downtime. This article is for anyone interested in reducing the backup window (improving the uptime of important applications) while backing up a system that is nearly online.
The process described here is called nearly online because you must take into consideration that to create a consistent snapshot of a file system, you must take the system offline to perform the following procedures:
Quiesce I/O on the volume.
Make sure the applications and data associated with this volume are in a good state (Oracle software must be in backup mode, file systems must be synchronized with utilities like fsck, and so on).
Postpone all write operations until the snapshot is complete.
While the backup techniques described in this article are not focused on specific versions of backup software or hardware, this article uses the following products to provide realistic examples:
Sun Solstice Backup_ softwareSun's OEM version of Legato's Networker software
Sun StorEdge_ Instant Image softwareSun's software snapshot product
Almost any supported version of these two products can be integrated into your networked computing environment to gain the advantages of reducing the backup window and backing up a nearly online system.
Why Create a Snapshot?
Creating a snapshot of a volume (file system or raw partition) has several advantages:
You obtain a PIT image of the data so all data is backed up at the same point-in-time, whatever the size of the master volume.
The snapshot relieves the original volume (master volume) from excessive I/O during the backup.
NOTE
I/O is only relieved on the master volume when the snapshot is an independent snapshot as opposed to a dependent snapshot (also referred to as a copy-on-write). When performing a read I/O on a dependent snapshot, unmodified blocks are read from the master volume and modified blocks are read from the shadow volume. By contrast, read I/O from independent snapshots is performed entirely from the shadow volume, relieving the I/O on the master volume.
The latter point is true in the case where the snapshot is a full copy (independent snapshot) of a volume. As a consequence, a volume that has been copied can be made available to the original application. This has the following advantages:
The window to backup this volume is very small.
You can obtain several backups during the day.
There is minimum disruption on the master volume.
A copy of the primary volume is available for special applications (testing, data warehousing, and so on).
Comparison
To appreciate the benefits of a PIT backup system, consider the comparison between a PIT backup and a non-PIT backup.
Traditional Backup
In a traditional non-PIT file system backup, the backup process works schematically as shown in the following figure.
FIGURE 1 Non-PIT Backup
The backup server contacts the client, and arguments are passed defining what to save.
The client daemon selects all the files, and transfers them back to the server.
PIT Backup
When using the snapshot technology for a PIT backup, the backup process works schematically as shown in the following figure.
FIGURE 2 PIT Backup
The backup server contacts the client, and arguments are passed defining what to save.
The client creates the snapshot image (quick update).
The client daemon selects all the files on the snapshot volume and transfers them back to the server.
Notice that there is an additional step consisting of creating the snapshot volume. What happens is, in fact, a quick update because only modified parts (disk blocks) of the master volume are updated. Therefore, the creation of the snapshot is very quick.
Results
The following two graphs illustrate the I/O measurements that occur during the two types of backups.
In the first graph, the backup is performed without an Instant Image snapshot. The master volume experiences a sustained high level of I/O activity.
The second graph reflects the I/O measurements while performing a PIT backup using a snapshot with Instant Image software. You can observe that in terms of I/O, there is a load on the master volume at the beginning of the backup, but the duration is short when compared to the other I/O graph. This short period of master volume activity corresponds to the quick update phase of the backup. Following this, all the I/O is diverted to the shadow volume, relieving the master volume.
FIGURE 3 PIT Backup
FIGURE 4 Backup I/O Statistics With Instant Image Software