This letter is in response to "D2D backup success depends on software" (see InfoStor, March 2004, p. 48).
The key advantages of disk-to-disk (D2D) backup are performance and reliability. Enhancing restore performance is particularly advantageous because restoration of interleaved (or multiplexed) data from tape is often significantly slower than the backup— usually too slow. How-ever, there's a way to improve multiplexing.
The main problem with multiplexing is that tape devices write in tiny chunks. At restore time, data is spread out over thousands of tiny chunks on tape. The sequential nature of a tape device forces it to go back and forth as it searches for the correct chunk to restore. A better approach is to write to tape in larger chunks. One way to achieve this is to have a memory buffer on the server that is writing to the tape drive. The buffer receives data from clients in parallel and stores the data in "virtual buckets" in the buffer. When a buffer bucket is filled, the server then commits the data in that bucket to the tape device. On tape, data is in large chunks and is sequential.
There are a number of advantages to this approach: First, it eliminates the problem of multiplexing. Second, performance issues are disassociated from their impact on the tape drive. A disk buffer eliminates this problem as well. Data flows to the buffer bucket first, and the tape drive is only used when a large chunk of data is ready to be written. A final advantage is that a buffer can maintain the backup data it holds even after it has been committed to tape, so that emergency restores can happen directly from the buffer—a similar benefit that D2D offers, but without the implementation challenges.
D2D offers many advantages that can enhance backup and restore. But tape will not go away. With smarter backup software many of the problems inherent in tape backup, including multiplexing, can be eliminated.
Senior Systems Engineer, Atempo