Historically, backups were classified as full or incremental backups. This is better discussed using the concept of backup levels, as are used in the traditional (though outdated) Unix backup utilities dump and restore. The levels begin at 0 and increase as the amount of information they back up is reduced:

Here full backup could mean a full backup of everything that is backed-up. It could thus be limited to a single partition or to a single directory such as /home. In the traditional backup terminology, a full backup is a level zero backup and all higher level backups are incremental backups.

Using backup levels you can greatly reduce the amount of information being backed up. The process, of course, relies on maintaining a file of time stamps of the most recent backup performed at each level. (In the case of dump/restore, the timestamps were kept in /etc/dumpdates.) 

Using this scheme we could implement a simple backup plan

• the first day of each month, do a level 0 backup. Create a timestamp to indicate the time and date the backup began. We will call this timestamp the level 0 timestamp.
• The next morning and each Sunday morning afterward, a level one backup is done, containing every file changed since the last level 0 timestamp. A timestamp is created when this backup begins (the level one timestamp).
• Each morning thereafter (until either Sunday or the first day of the month), a level 2 backup is done, containing everything that changed since the level 1 timestamp.
• As the month progresses, the size of the level one backup grows weekly. As the week progresses, the size of the level 2 backup grows daily.

A backup scheme such as this greatly reduces the amount of information that must be backed up daily. It also has the advantage of only requiring three active backups: the most recent level 0, 1, and 2 backups.

Note: complete backup utilities (such as dump and restore) created output that contained not only the files backed up but information on any file deletions as well to allow recreation of the current system state.

There are several disadvantages to this scheme

• the granularity is one day. The best that one could hope for is to restore a file in the form it was in the previous evening. This may not be sufficient for development projects.
• as the creation of backup levels relies on the timestamps, their correct creation must be ensured.

Restoring information

The method of restoring information from the backups depends on whether you are recovering the entire data set or simply a deleted piece. To recover the entire dataset, restore the backups from the lowest level to the highest level sequentially. To recover a single piece of data, look for it starting with the highest level backup sequentially to the lowest level. The highest level that has a copy is the most recent version saved.

These restore procedures illustrate the drawback of adding backup levels. Although it decreases the amount of data that must be backed up at the highest level, it makes finding a particular piece of data more difficult and delays the full restoration of a system.

A simple incremental backup plan can be implemented using one of the archive programs we will cover next, our friend find with the option -newer, and timestamp files created by touch. This plan does not account for data deleted if a full restore is required, however.

Copyright 2008 Greg Boyd - All Rights Reserved.