It is a straightforward extension to the simple physical backup described in the preceding paragraph to interpret the file system meta-data sufficiently to determine what disk blocks are in use and only back those up. All file systems must have some way of determining which blocks are free, and the backup procedure can interpret that information to only back up the blocks that are in use. Naturally, this requires that the block address of each block written to the backup medium be recorded so that restore can put the data back where it belongs.
The primary benefits of a physical backup scheme are simplicity and speed. It is simple because every bit from the source device is copied to the destination; the format of the data is irrelevant to the backup procedure. It is fast because it is able to order the accesses to the media in whatever way is most efficient. There are a number of limitations to physical backup, however. First, since the data is not interpreted when it is written, it is extremely non-portable; the backup data can only be used to recreate a file system if the layout of the file system on disk has not changed since the backup was taken. Depending on the file system organization, it may even be necessary to restore the file system to disks that are the same size and configuration as the originals. Second, restoring a subset of the file system (for example, a single file which was accidently deleted) is not very practical. The entire file system must be recreated before the individual disk blocks that make up the file being requested can be identified. Third, the file system must not be changing when the backup is performed, otherwise the collection of disk blocks that are written to disk will likely not be internally consistent. Finally, the coarse grained nature behind this method leads to its own difficulties. Because file system information is not interpreted by the backup procedure, neither incremental backups nor backing up less than entire devices is possible. A raw device backup is analogous to a fire hose. Data flows from the device simply and rapidly -- but it is all the you can do to hold the hose. Finer grained control is generally impossible.
These negative aspects of physical backup have until now caused it to have very limited application. Physical backup is used in a tool from BEI Corporation that addresses the problem of restoring NT systems after catastrophic failure of the system disk [Edw97]. Two large experiments at getting Terabyte per hour backup performance mention the use of raw (or device) backup and contain performance measures that validate the intuition that it obtains extremely good performance [CCC98,Sun97].