On 19 Dec 2002, Nathan J. Williams wrote:

> Bill Studenmund <wrstuden@netbsd.org> writes:
>
> > > > What do say Veritas or IBM's LVMs do?
> > >
> > > I have no idea.
> >
> > I'd suggest we do something like that.
> >
> > Actually, it would be VERY NICE to have the same disk layout as one or the
> > other LVM. :-)
>
> What's "VERY NICE" about it? Much of this exercise is about having
> tools to parse N formats of partition tables; once we can do that,

In my mind, part of it is, and part of it is about throwing all of the
partitions into a big fat pool and swirling them around. The former I like
a lot, the latter I don't.

I see the consequences of the swirling-into-one-big-pool as being very
negative, and not getting us benefits to offset them.

> then the layout of one or another LVM can just be handled by another
> partition-reader, and I don't see what the benefit is of using that
> format for our own purposes.

No. LVM logical volumes aren't wedges, and wedges aren't logical volumes.

Most of the things I want from an LVM can't be done with wedges. The big
concept is that a logical volume is made up of a set of extents, whereas
wedges are just single extents.

Among other things, the only logical volumes we could express (say from
reading an IBM/OSF/HPUX volume group) in terms of wedges are very simple
ones; ones where there is no striping or mirroring going on, and where the
whole LV was allocated in terms of one extent. Anything more sophisticated
than that won't be expressable via wedges.

While I think the kernel might need something like wedges in an LVM
subsystem, I don't see why we need to expose it to userland.

Take care,

Bill