1 |
Dude - I use xfs w/o a UPS for desktops and laptops. I use it on |
2 |
servers with RAID and with UPS protection. I also keep good backups |
3 |
for the servers. I have been using XFS since _just_ _after_ it came |
4 |
to Linux. I have used XFS on several hundred systems (which I have |
5 |
been responsible for). I have, to date, lost two filesystems. |
6 |
|
7 |
a. 2000 - I lost a filesystem when I was running a CVS kernel. |
8 |
*hhahaha* yeah, it was ugly, during the 2.4 kernel of pain days. |
9 |
|
10 |
b. 2006 - hardware slowly corrupted an FS. Some files wouldn't read, |
11 |
and we had wierd problems, (but good backups). After firmware updates |
12 |
problems got stranger, and xfs_repair finished the job. I blame the |
13 |
hardware. |
14 |
|
15 |
I have read the list, and seen the problems. I don' t know what I do |
16 |
that makes XFS succeed, but It really does work well. The first |
17 |
filesystem I ever tried with JFS failed. I had weird errors, and |
18 |
strange messages. I tried the repair tools, but they crashed. Then, |
19 |
I posted to the LKML. No one replied, or was interested. I left JFS, |
20 |
and returned to XFS. |
21 |
|
22 |
I have run into a few strange bugs with XFS, but in every case I found |
23 |
the mailing list and IRC very responsive and I was able to return the |
24 |
servers to operation. Twice those have been caused by either 2.4 or |
25 |
XFS. Once or twice it was several compound power outages. |
26 |
|
27 |
What really kills XFS is _NOT_ power outages - it is out-of-order |
28 |
commits. When the drives re-order the commits, it really can f-up the |
29 |
drive. The data portion of the disk is updated and the journal isn't. |
30 |
Then, if you have a crash, you are in some pretty sh*t. That's why |
31 |
write barriers are so important. |
32 |
|
33 |
Use what you want, but don't misunderstand XFS - as many people here |
34 |
clearly do. It's a good FS, but it is sensitive to hardware problems. |
35 |
By problems I mean: dying disks (which will kill anyone), faulty |
36 |
commit order for data vs. journal (which probably affects all of the |
37 |
journaling FSs as well), silent corruption, faulty RAM, and last but |
38 |
not least DMA problems. |
39 |
|
40 |
If you have a drive that commits out of order, and you are prone to |
41 |
power problems: USE EXT2 - it is , bar none, the SAFEST filesystem in |
42 |
that case. I do use it on a couple systems with those exact |
43 |
problems. (And a couple of low-memory systems, journalling sucks up |
44 |
resources). I have not lost an ext2 fs yet in either of those cases. |
45 |
Sure, the systems occasionally experience some meessed up files, but |
46 |
never the whole FS, and replacing a library or binary is /not/ /that/ |
47 |
/tough/. |
48 |
|
49 |
Good Luck, |
50 |
Joshua |
51 |
-- |
52 |
gentoo-user@g.o mailing list |