> "...
> However, any data changes that
> clients commit to the database after the backup begins are not
> recorded in the backup
> file.
> ...
> "

Martijn Tonies wrote:
" Well, that's a GOOD thing. How would a backup be a backup of
a particular point in time, if changes happened AFTER the backup
has started would be recorded?
"

Some backup methods allow to restore database to a particular point in time
which is not neccesserally a time when backup started but can be a later
timestamp instead. This is basically about maintaining a sequential change
log and it's backup which can be used to roll-forward a database after it is
restored from a main backup. The point is that change log is split into
rather small files and those files can be backuped as they are created.

Then to restore a database one needs to restore it from a main backup and
then roll-forward using backuped change logs:
- it takes time to roll-forward of course
- but one can restore database up to the last available change log file - if
log files are not too large then one loses as much data as fits into one log
file.

Basically this is about availability which can be achieved by other means as
well, e.g. by replication.







[Non-text portions of this message have been removed]