1i5t5.duncan at cox
May 10, 2012, 5:22 PM
Post #22 of 24
Lie Ryan posted on Fri, 11 May 2012 02:31:27 +1000 as excerpted:
> Also, it probably won't hurt using ext4, given that ext4 had several
> performance improvements over ext3. However, as Theodore T'so - the
> primary developer of ext4 - said, ext4 is a stopgap until btrfs is
> ready. IMO, from my experience of using btrfs some time ago, I'd say
> that btrfs is pretty much ready nowadays.
As long as you don't have problems or want to do anything fancy like
multi-disk, btrfs can be fine. But it's still in active development and
officially experimental, it only recently (Feb/Mar) got an error
correcting btrfsck at all, and that still comes with "it may make the
problem worse instead of fixing it" warnings. I'm running a few
partitions of it now, but as I tell people on the btrfs list, while a
good admin will always have backups no matter the stability of the
filesystem, with something as experimental and under development as
btrfs, it's best to consider your btrfs copy an extra "testing" copy,
that may or may not be there the next time you access it. Your primary
copy, along with all backups you'd ordinarily have, should still exist
and be located on something other than btrfs.
We routinely see people on the list asking how to recover data, because
they didn't heed that advice. Sometimes it's recoverable, sometimes part
of it is, sometimes not.
Just watching the commits and related discussion on either the btrfs
lists or as they hit the mainline kernel, they're still actively fixing
code broken in one way or another, as well as continuing to add
features. raid5/6 mode is roadmapped for 3.5 (some preliminary prep
commits went into 3.4), and full n-way mirroring raid1, the current so-
called raid1 mode is only two-way-mirroring) is roadmapped after that as
it builds on it.
But with a few more kernels, say by the end of the year or early next,
btrfs should really begin to stabilize.
Meanwhile, anyone who does choose to run it should be keeping up with the
latest kernels. If you're not running the rc kernels at least by rc5 or
so, you're running old code with known problems patched in newer kernels.
Duncan - List replies preferred. No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master." Richard Stallman