GNU Parted and LVMS, LVM, RAID, etc.

[Andrew Clausen]

Hi Ben,

I'm the GNU Parted maintainer.  GNU Parted is a partition tool.  The
big attraction is it can, among other things, resize partitions (and
the filesystems on them)  It supports:
* filesystems: ext2, fat, linux-swap
* disk labels: mac, msdos, pc98, raw device access

http://www.gnu.org/software/parted

Anyway, enough propaganda.

We have come to the conclusion that partitioning is a very hard problem
to solve well.  One feature (which I won't discuss much here) is
automatic partitioning, which is essentially knapsack problem, which
is NP hard (and before you ask, it IS large N, in common cases ;-)

Another big problem is allowing the user to plan multiple partition
operations in advance (like copying or resizing partitions), and
allowing the user to "commit" the plan.

The reason this is difficult, is the constraints that must be applied
to resizing.  For example, a partition will have a minimum size
(which is a function of, among other things, the amount of used space)
- so while doing the planning, one must keep track of how the file
system will be modified, and how the resize constraints change.

Representing this is non trivial, and designing an interface to do
this stuff seems like it's going to be some work... (what is an
"operation"? how does file sytem or disk label code communicate
changes of state?  or changes in constraints? etc.  REALLY HARD: how
does the file system code communicate, for example: "if you let me
change the block size down to X kb, I can now move the start of the
partition by any multiple of X kb")

Anyway, we are wondering how much of this kind of research IBM has
done already.  It's difficult to tell how sophisticated LVMS really
is, from the information that has been released so far.  Are these
problems in it's scope?  Do you see these issues as relevant to where
you're heading?

Also, we'd think a lot of Parted's functionality is useful for LVM,
LVMS, RAID, etc.  We think a unified interface, at some level, would
be a useful thing to have.  Is LVMS a good architecture to build
this with?  Is there more information about LVMS available, other
than the paper you wrote? 

http://oss.software.ibm.com/developerworks/opensource/linux/whitepapers/lvm/white_paper2.htm

Thanks,
Andrew Clausen