On Fri, Jul 07, 2006 at 11:13:44PM +0200, Paul Bijnens wrote:

> There are two approaches (maybe even many more -- I'm not the
> specialist here).
> 
> One approach is to let the snapshot mechanism understand the filesystem
> and work on that level. That is how xfs and solaris ufs snapshots seem
> to work.
> 
> The other approach is a layer deeper: on the logical volume manager.
> This implementation is not interested in where the inodes are located
> or how directories and datablocks are layed out on disk. The LVM just
> manages large blocks.  The LVM2 snapshots in linux work on this layer.

If the snapshot is on the driver layer, there is no way to ensure that
the snapshot is consistent?  Thus the snapshot would look like the power
switch was turned off at that moment.  Without journalling, data loss can
be expected.  I would rather not base my backups on such a mechanism...

> Using method 1, you can probably get more optimized (quicker, using less
> diskspace).  Using method 2 is more general, and is is independent of
> the filesystems on it.  In my test setup I even managed to make a
> snapshots of a vfat filesystem.

I don't think vfat have journalling.