On Mon, 2 Nov 1998 17:31:19 -0600 (CST), Shawn Leas
<sleas@ixion.honeywell.com> said:
> THIS is not the unix way. The unix way is to provide a primitave
> for userspace to do with as it needs. NOT make EXT2 like MS Word.
> Ted has done GREAT work in coding all that he has. He's the brain
> behind PowerQuest's Ext2 resizer in their Partition Magic program.
> However, I still have to take issue with shoving volume mgmt into
> the FS layer. Remember, LVM gives ALL of userland a simple block
> device to use. All the features, however you wanna use it.
This is not true.
There are at least two important features which *cannot* be achieved in
an LVM. One is online resize of filesystems. The second is filesystems
larger than 1TB.
Online resizing can be helped by a LVM, certainly. You can reduce the
general mapping problem into an issue of simply extending or shrinking
the filesystem by hiding the rest of the remapping in the LVM. However,
you simply cannot do online resize without filesystem support for it.
The 1TB issue is a hard upper limit on all block devices in the kernel.
I am *not* about to go making the entire block device level 64-bit
clean, with all of the auditing and rewriting of every single block
device driver in the kernel which that would require. 2^31 * 512 (the
sector size) is 1TB. We could conceivably push that to 2TB if we
audited the whole block device layer to be clean with respect to sign
issues, but we aren't going to increase it above that.
I won't try to pretend that these issues are killer obstacles for most
users: obviously they are not. But to pretend that they are not at all
important is simply blind. You need filesystem support here. There is
no question of that. Given that filesystem support for LVM is in the
long run a requirement, the main objection is that there's no point
doing a half-fix in the device layers when you will just have to redo
that fix in the filesystem eventually anyway.
LVMs are great. They just aren't a complete solution. The thing is, we
live in a free software world now. :) We can offer both and let users
work out what combinations make the most sense for them.
--Stephen
-
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/