I did some comparison of ZFS vs XFS last week to review the current state of the art in filesystems.

Long story short. XFS is still the reigning champion (at least on Linux). XFS beats out most filesystem benchmarks across the board. Reiser does well on directories with lots of small files but not enough to justify not using XFS.

Reiser FS is out of the picture honestly. First, it just doesn’t perform very well. Second, Hans Reiser is probably going to prison for murdering his wife and is selling the company to pay off his legal costs.

ZFS would have a shot on Linux if they could change the license to be compatible with the GPL. There’s a FUSE port of ZFS but honestly the performance of a userspace filesystem won’t ever be able to compete with XFS.

The native support of a volume manager in ZFS is pretty interesting though. That and copy on write semantics provide snapshots which are really important for some applications. I’d love to see XFS support snapshots but apparently it’s only supported by LVM.

Running MySQL on top of XFS with LVM and snapshots would be a great way to perform quick backups without having to have significant downtime. The short XFS freeze might not be ideal in some situations though. You’d still have to copy over the files and create a secondary copy which takes IO on the disk. This sort of removes any major advantage since for large backups the machine won’t be available for a long period of time.

  1. So did you actually do some benchmark comparisons between the two, or are you just calling this one out of the dark?

  2. It’s only prudent to mention that, regardless of other metrics, ZFS has no peer when it comes to data integrity (and is also next-generation in some other areas, such as manageability). Using any other filesystem, whether RAID or not, puts your data at greater risk.

  3. Why not simply create a SQL dump of the mysql database? oh, or can´t mysql guarantee consistent dumps?

  4. ishobo

    “I did some comparison of ZFS vs XFS last week to review the current state of the art in filesystems.”

    No you did not.

    – you only used Linux
    – you only used XFS since there is no ZFS for Linux
    – you did not post your methodology
    – you did not post your results

    “ZFS would have a shot on Linux if they could change the license to be compatible with the GPL.”

    The inverse is true as well. ZFS would have a shot on Linux if the license was changed from the GPL. What is your point?

    As has been dicussed on the zfs-discuss mailing list, porting ZFS is more technical than license. The porters could write an abtraction layer – a GPL module that provides an ABI for the CDDL code. If Linux folks wanted ZFS on Linux, and not via FUSE, the port would have been started by now.

    The title of this should have been, “My thoughts on ZFS on Linux, and XFS runs pretty darn quick in my mind.”

  5. I have noticed licensing issues are becoming more of a holding back progress and innovating new things lately. While I realize, outside of the academic world where I live in, like the real world, licenses are required to ensure things remain open and available to everyone, so that some big redmond based company does not swallow up everything.

    But consider a reverse situation, someone wanting to use an XFS filesystem from FreeBSD. As far as I know, no stable port exists. And they cannot just use the GPL licensed one that already exists because its license is incompatible with the BSD licensing model. (well it is also written for Linux, and would need to be ported),

    So because of license incompatibilities, to make an XFS port for FreeBSD would require a significant duplication of effort, that could have been allocated to improving or developing something else.

    If only these licenses could just all get along :)

%d bloggers like this: