On Wed, Apr 25, 2001 at 08:28:39AM -0700, Traitel, Eyal wrote:
Our snapshots are to be more exact "copy on new write" as opposed to "copy on old write", with the advantage of no penalty at all in terms of performance of taking the snapshot, and performance of filesystem after taking it (think of "copy on old write" - filesystem needs to read old block, write it in the backup location, then write the new block in old place - that's 3 I/Os. For us it's only 1). We do not have to go back to specific locations on the disk, and that's a real advantage in terms of performance.
In regards to full compared to snapshot - with NetApp, it's probably always better to backup snapshots, as they are point-in-time - when you've started your backup at 19:00, all of it will be consistent to 19:00 sharp, unlike traditional backup, which data changes while you backup... So - if you get me right - you can even do full backup from a snapshot...
There's one potential problem - if your backup does not finish in time between 2 snapshots, they will rotate (say, hourly.0 will become hourly.1, and the new hourly.0 has started) , so in the end you're backing up not what you started backing up. For scheduled backups one should be cautious to, for example, do a full backup from weekly.0 and incremental one from nightly.0, but if you're doing a on-the-fly dump, always keep your snapshot schedule in mind.
Igor