Archive for November, 2007

Why do we use tapes?

Tuesday, November 13th, 2007

I’m at LISA’07 in Dallas this week, and today attended Jacob Farmer’s training session, titled “Disk-to-Disk Backup and Eliminating Backup System Bottlenecks.” Early in the talk, he talked about how painful tapes are: the “batch” paradigm is a bad fit for object restores (the “I deleted a really important file! Please help!” problem), but doesn’t work terribly well for historical archives, either.

You can look at a number of modern backup configurations as ways of dealing with the desynchronization of tape cycles and retention periods. The problem is that tapes bind large amounts of data together in a single physical package, such that if *any* of that data is still required, the entire package must be retained. This poses a problem when an organization needs multiple distinct retention periods, or when different tapes are used for archival storage vs. object restores. The “synthetic fulls” technique gets around this by dynamically generating the necessary tapes, but this is obviously a brute-force approach that requires a lot of extra horsepower and time. The “incrementals forever” method, on the other hand, avoids tape altogether and exploits the random-access nature of disk to keep a consistent image without ever pulling new fulls from clients. It seems clear that disk is the answer here, and this is behind the popularity of D2D2T and related techniques.

In light of all this, it’s interesting that one of the more popular backup technologies is VTL — virtual tape libraries. These are boxes that take disks, which we’ve agreed are the way to go, and make them look like tapes. Their selling point, of course, is that they slot into existing systems with little or no change. It will be interesting to see whether these prove to be a “transition” technology, allowing organizations to migrate an existing configuration to a tapeless configuration without taking a sudden plunge.

The technical talks here at LISA’07 start today (Wednesday). I’ll be posting more more backup-related reflections here on the Zmanda blog, and will have some more general musings over at, my development blog. If you’re at LISA, please look me up at the Amanda Development BoF tonight.