On Thu, 6 Jan 2005 22:12:38 -0700
Collins Richey <[EMAIL PROTECTED]> wrote:

> That's encouraging to hear. 
> 
> My only 3 experiences with reiserfs (1x late 2.4, 2x on 2.6, never
> again in this lifetime) led to total fs corruption after a powerfail.

Power failure is completely different to a crash! Unlike in case of a
crash, internal disk buffers are not flushed in case of a power failure.
This means, that *every* journaling FS can fail completely. If everything
works well, it is pure luck. In common setups, journaling filesystems do
*not* protect against damage from power loss, not even those that do full
data journaling. This is something that many people are not aware of.

The reason is, that modern disks reorder write operations, meaning that
data arrives on the disk in a different order than the filesystem expects.

This is very dangerous. There are several solutions:
1) disable write caching via hdparm -W
(This can severly lower write performance and even shorten a disk's
lifetime)
2) use an UPS
3) use kernel 2.6.9 or later, which supports so called "write barriers",
at least in ReiserFS and ext3.
Run "dmesg | grep flushes", you should see something like
hda: cache flushes supported
hde: cache flushes not supported
Here hda suppports barriers, hde doesn't.

Unless you use one of those solutions, every journaling fs can, and some
day will, experience corruption on power failure.

There is a web page where some test results are displayed. They managed to
damage both ReiserFS and ext3 beyond recovery with only 3-4 power
failures.
Unfortunately I lost the URL...

Regards

--
[email protected] mailing list

Reply via email to