LifeBoat: An Autonomic Backup and Restore Solution
LISA '04 Proceedings of the 18th USENIX conference on System administration
A risk analysis of disk backup or repository maintenance
Science of Computer Programming
| Hi-index | 0.00 |
Every system administrator depends on some set of programs for making secondary copies of data, as backups for disaster recovery or as archives for long term storage. These programs, like all other programs, have bugs. The important thing for system administrators to know is where those bugs are.Some years ago I did testing on backup and archive programs, and found that failures were extremely common and that the documented limitations of programs did not match their actual limitations. Curious to see how the situation had changed, I put together a new set of tests, and found that programs had improved significantly, but that undocumented failures were still common.In the previous tests, almost every program crashed outright at some point, and dump significantly out-performed all other contenders. In addition, many programs other than the backup and archive programs failed catastrophically (most notably fsck on many platforms, and the kernel on one platform). Few programs other than dump backed up devices, making them incapable of backing up and restoring a functional operating system. In the new tests, crashes are rare, and more programs function as backups of entire systems, but failures in backup and archive programs are still widespread, and dump is not inherently more reliable than other programs.