As I've followed the backup market during the past few years, I've been impressed by how the pendulum has swung once again to favor the various flavors of disk image backups. With announcements from EMC and Veeam expanding the reach and functionality of changed block tracking, it seems images may become the dominant backup mechanism for the next few years.
Image backups are, of course, nothing new. In the early days of the PC LAN, backup systems from vendors like Alloy Engineering and Tallgrass Technologies backed up hard drives sector by sector to DC-300 data cartridges. The SCSI committee even built bulk copy operations into the common command set to empower image backups.
While these primitive image backup systems were, at least for their day, fast, they were problematical in many other ways. The biggest problem was that you could really perform only a full-disk backup, and, more significantly, a full-disk restore. When a user called and asked for a single-file restore, we would have to restore to a scratch disk and copy the file back. Once Backup Exec and ARCserve came around and made file restores easier, image backups were delegated to niches and some SMB users. After all, the vast majority of restores are for a small group of files.
Of course, the shift to file-by-file backups complicated the process of a full system restore. While full-system restores are rare, the very thought of an Exchange server suffering a catastrophic failure is enough to make most system administrators break out in a cold sweat.
As a result, a second generation of image backup software like Symantec System Recovery, which was until recently known as Backup Exec System Recovery, and Acronis Backup cropped up. With second-generation image backup, admins could do individual file restores and incremental backups. Even better, these programs could boot from a CD and automatically update a server's driver set during the restore process, vastly speeding the process of restoring a server to different hardware.
Today, the vSphere vStorage API for data protection lets backup apps take application consistent image backups of virtual machines by accessing vShpere snapshots of the virtual machine on shared storage. This takes the load of processing backup data off the virtual server host, eliminating a major bottleneck. If that weren't enough, it also does changed block tracking so subsequent incremental backups are smaller and faster than the old file-by-file kind.
With recent updates like EMC’s Avamar using changed block data to revert a VM back to a previous state, along with Veeam’s using VSS to take image backups and adding changed block tracking to Hyper-V environments, it seems there are few good reasons to back up systems file by file anymore.