[Gllug] Sony AIT Tape

[Mark Hemment]

On Tue, 17 Jun 2003, Tethys wrote:
>
> Richard Jones writes:
>
> >> Using dump on a mouted file system is, errr, "brave", to say the least.
> >
> >But sometimes necessary. How else do you reliably back up a 1/4TB
> >filesystem with a million subdirs on it over the network to an Amanda
> >backup server?  On a 24/7 service?
>
> With a snapshotting filesystem and/or volume manager, of course.
>
> >Linux really needs a way to do this properly ...
>
> See above. I believe recent versions of LVM support snapshotting,
> although I haven't played with it yet. If you've got big bucks to
> throw around, and you're on an IA32 Linux box, then you could look
> at Veritas Foundation Suite, which handles snapshotting (though
> they call it PIC -- point in time copies).

  A bit more detail...

  Snapshots are a function of the volume manager - it creates a
point-in-time mirror of a volume.  If the original volume contains a
mounted filesystem, then for a "good" snapshot the volume manager requires
the filesystem to flush the dirty buffers, inodes, etc, before the
snapshot is taken.  Once created a snapshot can be 'fast' re-sync'ed
against the original volume.

  Checkpoints, aka clones (aka PICs), are a function of the VxFS
filesystem.  VxFS has the concept of filesets, which (at one level) can be
thought of as multiple filesystems within a filesystem.  Creating a
checkpoint takes a "copy" of a 'frozen' image of the original
filesystem/set.  The checkpoint can be mounted just as any other fileystem
- even read/write!  VxFS attempts to share extends (data blocks) between
filesets - copy-on-write - this reduces block usage, and the time to take
a checkpoint.  Also possible to have checkpoints of checkpoints (fun!).
There are also "Nodata" checkpoints, but they're just weird.

  Then there are filesystem snapshots.  These are designed for fs
backups; read-only, good read throughput.  They need a separate storage
device (to store 'original' data as it changes on the primary fileset),
and (unlike checkpoints) are not persistent.


  One of these days, I'll do a gllug presentation on Veritas'
filesystem and volume manager - VxFS and VxVM - and try to explain why
they cost so much.

Mark



-- 
Gllug mailing list  -  Gllug at linux.co.uk
http://list.ftech.net/mailman/listinfo/gllug