[dm-devel] Re: Was: [patch 1/1] md/dm: Reduce stack usage with stacked block devices

[Mike Snitzer]

On 1/11/06, Ming Zhang <mingz at ele.uri.edu> wrote:
> On Wed, 2006-01-11 at 20:54 +0000, Alasdair G Kergon wrote:

> > Efficient, unlikely with the current implemention - hence the need
> > for a rewrite.
>
> >
> > If you have 5 snapshots of the same origin, the current implemention
> > takes 5 copies of the data when you change it. That approach is
> > never going to scale well!  And it uses lots of memory.
>
> yes, you are right. is there any new rewrite action proposed or planned?

My understanding is that much of RedHat's Cluster Snapshot Block
Device (CSNAP) core is a candidate for the dm-snapshot rewrite:
http://sources.redhat.com/cluster/csnap/csnap.ps

Along with other improvements, performance would be addressed with a
shared exception table (tree) for all snapshots (only one copy of each
exception).  When I spoke with Daniel Phillips (author of CSNAP
stuffs) a few months ago he said CSNAP is on the back burner until
others realize just how bad the existing dm-snapshot is.  He is
looking/waiting for others who are willing to contribute in moving
CSNAP's snapshot improvements into dm-snapshot++.  Unfortunately I'd
only be able to contribute testing at this time...

Mike