1 |
On Sat, Sep 6, 2014 at 4:54 PM, Kerin Millar <kerframil@×××××××××××.uk> wrote: |
2 |
> From reading the XFS list and my own experience, I have formed the |
3 |
> opinion that the maintainers are more stringent in matters of QA and |
4 |
> regression testing and more mature in matters of public debate. |
5 |
|
6 |
That doesn't surprise me. One of the best tools for QA testing any |
7 |
filesystem is xfs_test, which was, as is obvious from the name, |
8 |
developed to stress xfs. I know the btrfs devs use it heavily, though |
9 |
it doesn't test all the more modern features of btrfs like |
10 |
snapshotting, reflinks, send/receive, and so on. |
11 |
|
12 |
I know the whole lkml debate about data=ordered didn't thrill me all |
13 |
that much. I'm a firm believer that no filesystem should eat your |
14 |
data if it doesn't cleanly unmount. I don't have a problem with |
15 |
losing the last n seconds of changes because of write caching. What I |
16 |
do have a problem with is when after a crash a file contains something |
17 |
other than the previous contents or the new contents, especially if a |
18 |
failed append to a file ends up zeroing out the whole file or some |
19 |
nonsense like that. |
20 |
|
21 |
> It is also one of the few filesystems besides ZFS that can dynamically |
22 |
> allocate inodes. |
23 |
> |
24 |
|
25 |
FWIW, btrfs also dynamically allocates inodes. ZFS and btrfs are |
26 |
fairly comparable in terms of capabilities, with each now having a few |
27 |
features the other lacks. Btrfs is definitely less mature though. |
28 |
|
29 |
-- |
30 |
Rich |