Tape advantages
- Size: Tape backup up to 40 GB is available for < US$500 for the drive, and ~ US$15 for media. Compare to DVD as others recommend -- you're limited to 4GB. CDR is 650-700 MB -- at today's disk sizes, that's the equivalent of backing your hard drive onto 184 floppies (650 MB CDR and 120 GB drive). Systems of 200GB (single media unit) and more are available, tape 'bots extend this range further (all at a price).
- Cost: Incremental media cost is lower than any other format.
- Expandability: My current employer uses online disk backups, with no tape archive. While this makes for convenience, it means that extraordinary measures are required when space gets tight. You can't just "toss in another drive". The backup system's case is packed, so rather than drive install, it's a wipe the archive and build a new array. This makes for week or more's downtime as issues are worked through. I've seen this. Using tape, you add another cartridge to your rotation.
- Depth of archive. Disk inherently limits the ability to archive older copies of content. With tape, you simply rotate an archive out of active circulation, and it becomes permanent. The cost is your media cost per unit -- $15.
- Reliability: SCSI DDS technology has many years of high-reliability operation under itself. Problems with either tapes or drives are rare. Media themselves are largely impervious to shock damage, and can tolerate wide environmental extremes. The hardware is ubiquitous and standard -- if you lose your drive along with other hardware or data, you can still recover your tapes on COTS hardware elsewhere.
- Reusable media: individual tapes are rated for hundreds of duty cycles, and can be used far beyond this.
- Differential backups: to address the issue that tape tends to be just slightly smaller than the data you want to archive, a strategy of differential backups is highly recommended. Under this strategy, a periodic full backup is made (say, weekly). A differential backup is made of all changes since the nth most recent full backup, where n is greater than 1. That is, you're doing a differential from three or four weeks back, if you're doing regular weekly full backups. Why, you ask? Because this means you can now pair a differential with any of the past four weeks' full backups. You'd have to lose four weeks' worth of full backups before your incremental became useless. Note that the "tower of Hanoi" scheme is particularly fragile, as loss of any one backup in the "tower" results in loss of data. This was pointed out to me by Rick Moen. It was pointed out to him by that most useful of all System Administrator training methods, bitter experience.
Tried, true, tested, and cheap, hard to beat.
Note that backups to disk may make sense for a single, nearline, copy of data, and as a staging system for tape backups. However, I wouldn't (and don't) trust this as my long-term archival system.