opinion on XFS

Robert G. Brown rgb at phy.duke.edu
Tue May 7 11:15:08 PDT 2002

On Tue, 7 May 2002, Roger L. Smith wrote:

> On Tue, 7 May 2002, Yudong Tian wrote:
> We run XFS on the head node of our cluster on both the system partition as
> well as on a striped data partition.  I switched to XFS on it when my
> users were able to scramble the ext2 filesystem on the data
> partition almost daily.  At the time, the only answer was Reiserfs or XFS,
> and since I couldn't (easily) make Reiser the root filesystem, and since I
> had a lot of experience with XFS under IRIX, I decided to go with XFS.

Dear Roger and Eray,

I'm curious, how did users scramble the ext2 filesystem, and are you
certain that it was a weakness in the filesystem that was responsible
rather than the malfunctioning of some other kernel or system component
that was expressed as a filesystem corruption (a bad disk or controller,
for example)?  We've been using ext2 for years in a moderately demanding
client/server environment and haven't lost data due to a filesystem
problem since 2.0.1 (smp), and THAT was as much due to the fact that the
Adaptec SCSI controller I was using a) needed a bios flash and b) was
supported in linux without any help at all from Adaptec (a bit the
opposite of help, actually) as it was to ext2.  Aside from that one
incident I've never lost an entire filesystem, and all the times I've
lost files at all it was similarly due to a dying disk (bad blocks
appearing in two many key places on the disk) or broken drivers that
were causing the kernel itself to screw up not-quite-fatally.

Mind you, waiting for an fsck after a hard powerdown or crash sucked,
and yes, one COULD lose a file that was open and being written during a
crash, but crashes and power failures are relatively rare and even then
I almost never had to "do" anything to recover BUT wait.

ext3 does seem to make even those problem go away -- I don't actively
and deliberately turn a system in mid operation any more often than I
must, but my Dell Inspiron laptop had (really has) some odd BIOS
features that screw up shutdown attempts if one accidentally closes the
system while running, and with ext3 it just quietly comes back up.  More
broadly, we've had no serious problems reported at Duke.

> I've never had a problem with XFS in the nearly 18 months that I've used
> it.  However, I've also done testing with ext3 and am fairly satisfied
> with it, and the users no longer use the data partition on the headnode,
> so I can't justify the hassle of running XFS anymore.
> I'll probably switch the head node to ext3 in the near future.  However,
> if XFS were more formally supported by the kernel/RedHat, I'd seriously
> consider switching all of my Linux systems to it.

It's always nice to have a choice, but it is also nice to have a
functional default that just "works".  A good filesystem should be
basically invisible, no matter which one you use.


Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu

More information about the Beowulf mailing list