[Discuss] ZFS vs. Btrfs

[richard.pieri at gmail.com (Rich Pieri)]

On Wed, 02 Jan 2013 03:01:23 -0500
Tom Metro <tmetro+blu at gmail.com> wrote:

> You started out using ZFS on that server, right?

Indeed, I did.

> What were your reasons for switching to Btrfs and how have your
> comparative experiences been?

The primary reason is performance. ZFS with FUSE has a lot of CPU
overhead and the cores in the N40L aren't beefy monsters to begin with.

The secondary reason is that Btrfs is going to become the default file
system for Linux in the foreseeable future (it already is for some
distributions) and I figured it would be a good opportunity to try it
out.

I cannot honestly compare performance since I was running RAID-Z with
ZFS and mirrored data+metadata with Btrfs. I have a test computer at
work that I will use at some point to make valid comparisons.

Btrfs has all of the critical features of ZFS. It lacks some of the
features like independent ZIL, send/receive, and data+parity (RAID-5).
Compression is still all or nothing within a Btrfs pool. And there are
some rough spots with the non-Btrfs user space tools (df shows raw
storage rather than usable storage). These are coming but development
has been slow, it seems.

Btrfs is better in some ways. Devices can be removed and volumes can be
shrunk, a few others that ZFS users have wanted for ever but never got
from Sun. This isn't useful to me for my home server but it is very
useful for workgroup and enterprise projects.

Btrfs handles removable devices just fine. Mount and unmount like any
other file system. ZFS... crashes if you don't properly export the pool
first. Kind of a pain, really.

I've not ruled out migrating back to ZFS. I previously mentioned using
Debian kFreeBSD. That should be a best of both worlds: Debian user
space with native ZFS in the BSD kernel.

-- 
Rich P.