[linux-lvm] What have snapshots got to do with LVM ?

[Joe Thornber]

On Mon, Apr 23, 2001 at 12:05:21PM +0000, Heinz J. Mauelshagen wrote:
> On Mon, Apr 23, 2001 at 09:38:18AM +0100, Joe Thornber wrote:
> > On Mon, Apr 23, 2001 at 10:21:45AM +0000, Heinz J. Mauelshagen wrote:
> > > On Sun, Apr 22, 2001 at 01:09:24PM +0100, Joe Thornber wrote:
> > > > Just a random Sunday morning thought:
> > > > 
> > > > Why don't we split snapshots off into their own driver so we can
> > > > snapshot any block device, putting the COW tables on any other block
> > > > device ?  
> > > 
> > > Any other block device devides into PE similar allocation units carrying their
> > > own COW exception table at the beginning?
> > 
> > I'm not sure I understand your comment.  There are two devices the
> > snapshot origin and the snapshot storage device.  We can divide the
> > snapshot storage device up how we want, the COW exception tables go in
> > here.
> 
> With LVM sanpshots we devide a device into PEs, allocate parts of the
> COW table dealing with exceptions stored in a specific PE at the very
> beginning of the PE.

Yes, we would divide the snapshot storage into chunks consisting of
COW table + data.  This chunk size would be configurable, so we can
make it the same as the PE size *if* we were taking a snapshot of an
LV.

> My point is, that a similar allocation policy needs to take place on a
> general block device too in order to avoid disk access overhead.

Still not getting it.  Am I being stupid ?

- Joe