Virtual server backup is an increasingly important process in today's data centre. But what is actually copied during a virtual server backup and what is restored during recovery?
When taking an image-level backup, a snapshot is taken of the virtual server. For this to happen, all buffers within the server are flushed to ensure write operations are completed and that the backup is consistent. A copy -- or snapshot -- of the virtual disk files is then created and mounted to a staging location available to your backup server.
After this copy has been mounted to the staging location, it is backed up to your choice of backup device. If you restore from this snapshot, you get the virtual disk files as they were at the point of the snapshot. This gives you a complete disaster recovery image of the virtual server, which can be restored directly into the source virtual infrastructure or imported into a new infrastructure to give you a brand new server (with the same host name, IP address, etc., as the source server). In such a case, the restore process is an all-or-nothing affair in which you restore the entire server and all of the server's data.
When you take a file-level backup, the virtual server will again have a snapshot taken of it, but this time the volumes within the virtual machine (VM) are discovered and mounted to the staging area.
A file-level backup allows you to restore individual files and folders to a staging area. Or, with a backup agent installed in the virtual server, you will in most cases be able to restore directly back into the server. The data is catalogued in the backup product in the same way as a physical server's file system backup is catalogued, and users will be able to browse files and folders on the virtual server for restore as if it had been backed up traditionally.
But file-level backups have their limitations. In most cases, operating system files or locked files will not be backed up, and data discovered in the virtual volumes is not structured. That means backup will not be application consistent. Therefore, this backup type is not suitable either for restore in the event of a disaster or for granular restores within applications such as databases, email, etc.
The key lesson is that to get the same level of protection and restore functionality you get from traditional backup, you effectively have to perform two backup operations. Most major backup products offer single-pass virtual server backups that give both backup types, but you should verify that this functionality is present before considering an investment in a new technology.
This was first published in May 2010