BTRFS - Opinions so far

Latest response

I am curious how many people are early-adopters of BTRFS.


How do you like it thus far?

Have you seen issues?  If so, how signficant?

Have you done benchmarking?  What were the results?

Had you used a simliar filesystem/volume/disk management suites previously (like Oracle ZFS)?


I like it a lot so far.  Development seems to be stalling a touch though.  Would like to see dedupe pushed harder.  It may not suit everyone to dedupe, but I can guarantee it would be a hit in my datacentres.  Especially on the back-end of the growing virtual desktop environment (all identical).

Also hierarchical storage management would be a killer.  Home users would benefit from a simple SSD+HD configuration, whereas datacentre storage managers could expand this right up to use the full range of disk speeds/costs available to them.  The ability to make blocks "sticky" at a certain level would help.  i.e. make sure the boot-up disk blocks stay on fast storage, even though they may be seldom used.


Early on I saw performance issues with filesystems with many small files. (to be fair, my test box didn't have many large files).  Didn't benchmark anything as it was on home kit, but the speed-up when returning to ext4 was visible.


Never used ZFS or the like before; have always juggled SCSI devices, LVM and multiple mount points to manage my data.  The capacities of extX have always grown fast enough to cope with the data volumes/types I've been involved with.


I tried btrfs first on my server and besides it was unusably slow
causing false timeouts on GDB testsuite it was even corrupting its data, I had
to immediately switch back as no single nightly testsuite runs were usable:

I wanted to switch back from btrfs but I could not despite having enough
space; I had to split my RAID1, do full mkfs, cp and RAID1-rejoin then:
    btrfsctl -r: No space left on device
    btrfs: No space left on device with 21GB available

And now I am left with btrfs on my notebook with Intel SSD and I have to use
it only as dumb terminal and do all my work on 3+ years old server with HDD
and ext4 because on SSD notebook the under-sec operations take even more than
10 minutes, I do not have so much time to wait for linking GDB each time.
    btrfs: poor performance

There is even simple reproducer for HDD ext4 2 sec -> SSD btrfs 1 minute test.
30x slowdown is not so horrible as the practical use with 10-15 minutes lock
ups but I have at least a reliable standalone reproducer for it.

I am going to re-mkfs the notebook to ext4 with F18 release as in the year of
waiting none of the fatal problems I filed have been fixed or even responded.