Subject: Re: Logical Volume Management for NetBSD
To: Jonathan A. Kollasch <jakllsch@kollasch.net>
From: John Nemeth <jnemeth@victoria.tc.ca>
List: netbsd-users
Date: 08/16/2006 23:13:09
On Jan 6,  5:40pm, "Jonathan A. Kollasch" wrote:
} On Wed, Aug 16, 2006 at 03:15:09PM -0700, John Nemeth wrote:
} >      Probably need to write it from scratch.  It may need changes to
} > the disk drivers (i.e. the concept of partitions goes out the window),
} 
} How can that work?  Non-contiguous area?

     Yep.  My experience is with HP-UX, not AIX; but, they are probably
pretty similar.  In HP-UX a physical disk is mapped to a PV (Physical
Volume).  It is then divided up into 4MB chunks called extents.  These
chunks are then assigned to LVs (Logical Volumes).  You can assign (or
unassign) any number of chunks to any LV at any time.  This allows you
to dynamically resize the "disk" underlying a filesystem.  On HP-UX, if
you purchase the Veritas filesystem, you can resize the filesystem
itself on the fly.  With HP's filesystem, you have to unmount it and
run a program to expand it (you can't shrink it).  If you create
several filesystems then expand one in the middle, it will have
non-contiguous space on the physical medium.

} > and will require an intermediate layer between the filesystems to map
} > logical volumes to physical disk space.  To do it properly will also
} 
} We've had a layer like that since 3.0 with dk(4).

     I rather doubt dk(4) will do the job (see above), since all it
does is map to a native "partition".

} > require filesystem changes to enable resizing on the fly.  If you've
} > ever used a real LVM, you will know that it is a pretty complex beast.
} 
} Yikes!

     Yep, it would be a lot of work to create a true LVM system.  But,
it would probably be worth it.  It would be a very impressive feature
to have and would open a lot of doors.  Even Sun didn't have it at the
time Solaris 10 was initially released (I believe ZFS has the
functionality).

}-- End of excerpt from "Jonathan A. Kollasch"