On 7/20/06, Cliff Wells <[EMAIL PROTECTED]> wrote:
As a more useful bit of info than anecdotes and scaremongering, here's a
decent article that covers XFS in fair detail and compares a few of its
major differences from the other journaled filesystems:

http://www-128.ibm.com/developerworks/library/l-fs9.html

Posting links to articles written by Gentoo's founder is cheating! :-)

But in fact, Daniel didn't really address the real-world reliability
of the filesystems.  He addressed it theoretically, and only in
relation to reiserfs, not ext3.  But in fact I disagree with one
assertion that Daniel makes:

"but writing metadata more frequently does encourage data to be
written more frequently as well"

In it's default configuration, XFS avoids writing data out to disk
until it absolutely has to, or a *significant* amount of time has
elapsed.  Only by tweaking /proc settings have I gotten it to flush
out data in a reasonable amount of time.

Are you seriously telling me that in all the years you have run XFS
filesystems, you have never seen /var/log/messages get padded with
nuls?  That is the kind of "corruption" that XFS is well known for.
(BTW, I _know_ this is a security feature.  But the fact is that ext3
users pretty much _never_ see this kind of data, um, "security").  It
may be great at maintaining it's own consistency, but it seems
particularly predatory to the files contained within it.  I've already
mentioned a recent corruption I had with XFS on one of my systems...

Besides, every time this discussion has come up here, the majority of
particpants have agreed that ext3 is the least likely to corrupt data.

Don't get me wrong.  I like XFS, and I am running it on my laptop and
desktop systems.  However I have tweaked the settings so that it
behaves like I want, and am very cautious about just hitting the
reset/power button when I get a lockup.  I have learned that the hard
way.

And my "bottom line" is: if someone came here and asked "what is the
most reliable filesystem", my answer would be ext3.  Hands down.

-Richard
--
gentoo-user@gentoo.org mailing list

Reply via email to