next up previous
Next: Carelessness backups Up: Supporting backups Previous: Calamity Backups

Archival backups

Archival backups are targeted at restoring some or all of an individual user's home directory, possibly after an account has been closed and deleted.

Due to the long term nature of archival backups it is not appropriate to tie them to a particular filesystem format, and so a scheme similar to the Calamity backups would not be appropriate even if it were practical.

A cpio or tar archive of the home directory is likely the best option for these backups. The main burden that this imposed on a filesystem is the fact that files are normally collected sequentially and there is no opportunity for read-ahead between files. A better approach would involve submitting read requests for multiple files and directories in such a way that throughput is valued over latency, and the order of the returned data is not important. Then the filesystem could submit large number of requests to a device before waiting for any of the data, and so provide lots of opportunity for re-ordering and grouping requests and thereby improve throughput.

It may be possible to use the new asynchronous IO primitives in 2.6 to achieve this improved thoughput. If not a special purpose interface to the filesystem will be provided to provide optimal access for backups.


next up previous
Next: Carelessness backups Up: Supporting backups Previous: Calamity Backups
Neil Brown 2003-02-06