7 ms·

Why not use ZFS on Linux?

asveikau 7y ago
It will never be in-tree, short of an act by Larry Ellison. Depending on an out-of-tree module can get pretty crappy on Linux. The upstream kernel will break you a lot. If your distro is on top of keeping your out-of-tree feature working then great. It gets messy whenever that is not the case and/or suddenly you want a driver or feature only available in a newer kernel.
- dbuder 7y ago
  Filesystem and Distro/OS are part of one choice imo.
- p_l 7y ago
  Larry Ellison can't relicense ZoL - they'd have to start from scratch porting the code from Solaris
  - _-david-_ 7y ago
    I think people typically mean that Larry Ellison could come out and say there are no issues with incorporating CDDL code into a GPLv2 code base. Ideally in a document signed by him and lawyers at Oracle given to the Linux Foundation. If that happened I don't think people would have an issue adding ZoL to the kernel.
    - p_l 7y ago
      There are no issues on the side of CDDL anyway. Larry can't change the fact that GPL is the source of incompatibility - the general interpretation is that GPLv2 forbids code that has stricter licensing terms than GPLv2, and CDDL happens to have that - in form of patent grant. It's also why GPLv3 code has a license issue with Linux kernel, which is GPLv2.
      - hvis 7y ago
        They can relicense Oracle-owned code under GPL. As just one example. There are probably other options, too.
        p_l 7y ago
        Yes, and it would require starting the port from zero, as I have said.
        hvis 6y ago
        Nah, it's the same code. There would be no need to create a new derivative work. It's a pipe dream anyway, of course.
        p_l 6y ago
        It's not the same code - OpenZFS has substantially diverged from ZFS v28 which was last common version between OpenZFS and Solaris, that's one thing. Another is that all the porting work in ZoL that isn't already GPLv2 is CDDL - and Oracle is not the copyright owner. In fact, the copyright owners would not want to change the license to GPLv2, unless maybe it was done as dual license.
    - asveikau 7y ago
      I was being a bit flippant rather than literal with the Ellison comment. I think I read a similar joke on an LKML thread or similar once.
mikece 7y ago
Opinions and dogma about licensing is one reason as well as a sense that ZFS isn't actually better than ETX4. I'm not saying I agree with that, only that it's why ZFS and Linux don't mix from what I can tell.
- tw04 7y ago
  It would appear canonical disagrees.
- _emacsomancer_ 7y ago
  ZFS and EXT4 have different use cases.
  - rkeene2 7y ago
    ZFS and EXT4 on LVM have a lot of overlap. A decade or so ago I made a comparison between ZFS, BtrFS, and EXT4 on LVM [0] that may be of interest. [0] https://rkeene.org/projects/info/wiki/BtrFS https://rkeene.org/projects/info/wiki/BtrFS
    - _emacsomancer_ 7y ago
      The checksumming/error-correction of ZFS is big difference, and the main reason I moved some of my file systems to ZFS (even with versioned backups, on EXT4, I suffered data loss due to corruption/bitrot). On the other hand, for a simple set-up on a lower resource system, EXT4 has lots of advantages.
      - rkeene2 7y ago
        A good, controversial read is available at [0]. I don't agree with his conclusions entirely, but the data he presents is interesting. And also paired with [1] to give space overhead and MTBF figures for various RAID configurations (where RAID-Z ~= RAID5, RAID-Z2 ~= RAID6) [0] https://www.jodybruchon.com/2017/03/07/zfs-wont-save-you-fancy-filesystem-fanatics-need-to-get-a-clue-about-bit-rot-and-raid-5/ https://www.jodybruchon.com/2017/03/07/zfs-wont-save-you-fan... [1] https://rkeene.org/projects/info/wiki/221 https://rkeene.org/projects/info/wiki/221

It will never be in-tree, short of an act by Larry Ellison. Depending on an out-of-tree module can get pretty crappy on Linux. The upstream kernel will break you a lot. If your distro is on top of keeping your out-of-tree feature working then great. It gets messy whenever that is not the case and/or suddenly you want a driver or feature only available in a newer kernel.

dbuder 7y ago
Filesystem and Distro/OS are part of one choice imo.
p_l 7y ago
Larry Ellison can't relicense ZoL - they'd have to start from scratch porting the code from Solaris
- _-david-_ 7y ago
  I think people typically mean that Larry Ellison could come out and say there are no issues with incorporating CDDL code into a GPLv2 code base. Ideally in a document signed by him and lawyers at Oracle given to the Linux Foundation. If that happened I don't think people would have an issue adding ZoL to the kernel.
  - p_l 7y ago
    There are no issues on the side of CDDL anyway. Larry can't change the fact that GPL is the source of incompatibility - the general interpretation is that GPLv2 forbids code that has stricter licensing terms than GPLv2, and CDDL happens to have that - in form of patent grant. It's also why GPLv3 code has a license issue with Linux kernel, which is GPLv2.
    - hvis 7y ago
      They can relicense Oracle-owned code under GPL. As just one example. There are probably other options, too.
      - p_l 7y ago
        Yes, and it would require starting the port from zero, as I have said.
        hvis 6y ago
        Nah, it's the same code. There would be no need to create a new derivative work. It's a pipe dream anyway, of course.
        p_l 6y ago
        It's not the same code - OpenZFS has substantially diverged from ZFS v28 which was last common version between OpenZFS and Solaris, that's one thing. Another is that all the porting work in ZoL that isn't already GPLv2 is CDDL - and Oracle is not the copyright owner. In fact, the copyright owners would not want to change the license to GPLv2, unless maybe it was done as dual license.
  - asveikau 7y ago
    I was being a bit flippant rather than literal with the Ellison comment. I think I read a similar joke on an LKML thread or similar once.

mikece 7y ago

Opinions and dogma about licensing is one reason as well as a sense that ZFS isn't actually better than ETX4. I'm not saying I agree with that, only that it's why ZFS and Linux don't mix from what I can tell.

tw04 7y ago
It would appear canonical disagrees.
_emacsomancer_ 7y ago
ZFS and EXT4 have different use cases.
- rkeene2 7y ago
  ZFS and EXT4 on LVM have a lot of overlap. A decade or so ago I made a comparison between ZFS, BtrFS, and EXT4 on LVM [0] that may be of interest. [0] https://rkeene.org/projects/info/wiki/BtrFS https://rkeene.org/projects/info/wiki/BtrFS
  - _emacsomancer_ 7y ago
    The checksumming/error-correction of ZFS is big difference, and the main reason I moved some of my file systems to ZFS (even with versioned backups, on EXT4, I suffered data loss due to corruption/bitrot). On the other hand, for a simple set-up on a lower resource system, EXT4 has lots of advantages.
    - rkeene2 7y ago
      A good, controversial read is available at [0]. I don't agree with his conclusions entirely, but the data he presents is interesting. And also paired with [1] to give space overhead and MTBF figures for various RAID configurations (where RAID-Z ~= RAID5, RAID-Z2 ~= RAID6) [0] https://www.jodybruchon.com/2017/03/07/zfs-wont-save-you-fancy-filesystem-fanatics-need-to-get-a-clue-about-bit-rot-and-raid-5/ https://www.jodybruchon.com/2017/03/07/zfs-wont-save-you-fan... [1] https://rkeene.org/projects/info/wiki/221 https://rkeene.org/projects/info/wiki/221