-   Linux - General (
-   -   BRTFS rational question, for the BTRFS crowd (

kschmitt 02-14-2010 04:33 PM

BRTFS rational question, for the BTRFS crowd
I was reading some articles on BTRFS today, and it became clear, the scope of this FS is pretty broad, far more than say, ext2/3/4, xfs, etc.

So I'm wondering, what's the rational for including so many things that are provided by the kernel, LVM and other standard tools? I'm not bashing or trolling, It's an honest question.

The two big examples of what I mean would be:
BTRFS does snapshots, so does LVM.
BTRFS does striping and mirroring, so does the kernel, and LVM.

Is BTRFS meant to be both a FS and an LVM replacement?

Just a little confused.


salasi 02-14-2010 06:13 PM

I'm not sure if this is any kind of answer for you, but compare not with

far more than say, ext2/3/4, xfs, etc
but Sun's ZFS.

At least they are both driven by the CoW mechanism, and ZFS does do a combined MD/LVM/fs replacement, end-to-end, job. As a result, it seems to be possible for ZFS to do stuff that is beyond the traditional linux file systems, like guaranteeing end-to-end data integrity and simplified disk/array management.

Note one error in the paragraph above is that filesystems themselves don't actually do anything. It is only in combination with the various support utilities that anything is done, so when I wrote "possible for ZFS to do stuff" I don't really mean that at all; I meant "the filesystem in combination with the drivers and support utilities". But still, at least superficially, ZFS is a very impressive piece of work.

Now an interesting, and maybe irrelevant thing: the chief architect for BTRFS, Chris Mason, works for Oracle. And what little morsel has Oracle been buying for themselves? Sun. So, Oracle have been supporting a guy working on providing a ZFS competitor for Linux, whilst buying the original, which they should have the possibility of GPL'ing. I doubt that this has any immediate impact, but you do wonder whether Oracle now regret what they have been spending their money on. Maybe, if you are as minted as Oracle, you never notice the small change that you spent on a top developer doing this stuff.

Quakeboy02 02-14-2010 07:11 PM

It seems like they're over-reaching on this one by duplicating services already provided by the kernel.

syg00 02-14-2010 08:42 PM

I disagree - the support tools for a filesystem implemented on a LV are a can of worms. Look at something as simple as resizing. What about removing a PV. What about data integrity ...
Much better to do it in the filesystem itself. Similar to zfs - which Linus won't allow unless the license changes.

I don't use BTR yet, but I will be.

salasi 02-15-2010 04:31 AM


Originally Posted by syg00 (Post 3863797)
Similar to zfs - which Linus won't allow unless the license changes.

True, but you can try out ZFS via FUSE. Don't know what performance impact that has, though.


I don't use BTR yet, but I will be.
Well, I am not prepared to commit to it until I have a better idea of how the performance numbers look (err, confusing is the usual answer), but I expect to avoid ext4 and jump to BTRFS, when the time is right. Although if BTRFS is much delayed in getting out what I consider an adequately stable version, or the perf numbers look worrying, maybe I'll reconsider.

All times are GMT -5. The time now is 07:46 PM.