[Lxc-users] nilfs

[Ulli Horlacher]

On Fri 2012-03-02 (01:39), Iliyan Stoyanov wrote:

> I'm currently using btrfs raid 1 for a production server with 4 LXC
> containers (SL6.x) on it (old single core opteron w/ 4GB ECC RAM). The
> host is Fedora 16.

I have tested btrfs on a standard Ubuntu 10.04.3 and one with kernel
2.6.38-13-server (backport). Both lead to a fatal kernel loop when doing
rsync of some GB: /var/log/kern.log got filled up until the filesystem was
full, while no program was responsive any more. Had to power off the system.
The developer of btrfs says, it ist still in a "highly experimental
state". Indeed.


> Could you share some of your observations of the nilfs 

I operate it only in a test environment with a test LXC VM and did some IO
tests and benchmarks (like the rsync above): everything works fine, smooth
and fast. Yesterday I deleted unintentionally the LXC container. I really
appreciate the snapshot feature :-)


> and why do you think it could be beneficial for LXC, besides snapshots,

Snapshots. That's it.


> as those can be done with both LVM and btrfs at this point.

LVM needs an extra partition for each snapshot, whereas in NILFS it is
just a subdirectory.
And I dislike to have an extra IO layer to have snapshots.
On other systems like Solaris/ZFS or ONTAP/WAFL, snapshots are an
integrated feature of the filesystems. Makes the handling much easier.
Less complexity is always good, it eliminates potential errors and
problems.

btrfs is not stable enough. See above.

-- 
Ullrich Horlacher              Server- und Arbeitsplatzsysteme
Rechenzentrum                  E-Mail: horlacher at rus.uni-stuttgart.de
Universitaet Stuttgart         Tel:    ++49-711-685-65868
Allmandring 30                 Fax:    ++49-711-682357
70550 Stuttgart (Germany)      WWW:    http://www.rus.uni-stuttgart.de/
REF: <1330645169.21101.11.camel at tablet>