> 1) Is MD the *clean* way to do what MD does?
> 2) Is MD cleanly implemented.
>
> I guess I think it fails test #1, but passes #2. I don't think
> this sort of messiness really belongs wholely in the kern. I like
> the idea of having the virtualization of an LV in kernel. It allows
> userland to do it's thing.
If you could swallow up the partition code I might agree with you,
because that would simplify the overall system. Without that, you
just add complexity for little gain.
> Kind of like KGI/GGI. Uhh... Don't mean to start another holy war...
Old threads never die.
>> show me some documentation/standard that specifies how LVM should look
>> like, apart from vague assertions like "uhm, something like Veritas or
>> AIX's stuff, or maybe HP's?".
>
> A layer that provides a virtual view of physical disk. One that
> has VGs [Groups of physical volumes (disks or partitions)], PEs [Extents
> of said VGs in tunable sizes], and LVs [Logical Volumes] which are made
> from PEs.
That would be slightly useful, but you can get most of that with MD.
Look at all the terms and tools you had to invent. MD is simple.
Just create 14 little partitions and join them up with MD as desired.
Unless you have some really compelling advantage, KISS.
> A volume group can have alocation policies [strictly contiguous, whatever]
> so you can control just how your disk pool, or, VG will allow use.
With MD, my policy can be "use everything except the corrupt part in
the middle" or "use the first and last bits because I want to torture
test my system".
> Because LVM is basically at it's core a physical -> logical address
> remapping mechanism, RAID, mirroring, and all the goodies fit REAL
> nice in userland.
>
> Wouldn't you like to push all that crap out to userspace?
If you are serious about RAID in userspace, go away. I'll assume you
didn't really mean to type that.
It might be nice to put partition code in initdata+userspace,
with the main kernel only using a mapping that was provided.
> Both Sun and HP use Veritas's LVM.
> (http://www.veritas.com/product-info/vm/index.htm)
>
> Heinz's LVM is concepually like it, I think.
> (http://linux.msede.com/lvm/slides.html)
If you want to write something like Veritas's LVM, make it compatible
at the physical disk level so that we can read foreign data. Either that,
or be compatible with NT. The PC partition system is going to be eliminated.
We don't really need our own incompatible LVM.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/