Announcement

Collapse
No announcement yet.

ZFS File-System On Linux Keeps Marching Along

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    Originally posted by highlandsun View Post
    ZFS looks pretty unusable to me. That was ZFS-FUSE though, so its performance was as slow as NTFS (FUSE). Nobody could ever get any serious productivity using that as a primary FS. I'd give the native stuff a try, but so far there's still nothing better than JFS...
    The premise of ZFS was never that it was blazing a blazing fast high-performance file system.

    The premise of ZFS is scalability, manageability, stability, data security, and features.

    ZFS would be great as a NAS, a media server (with big files such as movies), and file system for backup and archival.

    Comment


    • #17
      Originally posted by uid313 View Post
      The premise of ZFS was never that it was blazing a blazing fast high-performance file system.

      The premise of ZFS is scalability, manageability, stability, data security, and features.
      A lot of data centers and database-driven web servers rely on high performance I/O, not to mention file servers with multiple heavy writes (e.g., in the enterprise). Of course Solaris didn't intend to forgo these market segments so Sun's solution was to match up ARC and L2ARC with ZFS to dramatically improve performance. This might not show much in a home environment or on benchmarks, but like you said, it shows up when it's scaled.

      Comment


      • #18
        Originally posted by highlandsun View Post
        ZFS looks pretty unusable to me. That was ZFS-FUSE though, so its performance was as slow as NTFS (FUSE). Nobody could ever get any serious productivity using that as a primary FS. I'd give the native stuff a try, but so far there's still nothing better than JFS...
        ZFS via FUSE != ZFSOnLinux ("native" Kernel Module)
        And FUSE's module performance != native kernel module performance

        And that if we talk about non-tuned-for-specific-workload kind of performance.

        Originally posted by johnc View Post
        A lot of data centers and database-driven web servers rely on high performance I/O, not to mention file servers with multiple heavy writes (e.g., in the enterprise). Of course Solaris didn't intend to forgo these market segments so Sun's solution was to match up ARC and L2ARC with ZFS to dramatically improve performance. This might not show much in a home environment or on benchmarks, but like you said, it shows up when it's scaled.
        Well, the difference is not that BIG (you quoted @highlandsun that was talking about FUSE performance, which is crap) ... but with proper gear and setup you have the ability to have massive performance that other filesystems can't deliver, simply because they don't offer the features to make that happen.
        If you can spend on a SSD or two, you could use them as cache and/or ZIL devices ... and/or the natural raid setup with ZFS, which simply outperforms whatever black magic you can do on top of a "primitive" filesystem like ext4

        Again, ZFS is not intended for low-end desktop computers (think: ZFS for normal full operation needs at least 4GB .. with an optimal "start" of 8GB) ... so it's basically bogus to compare it against ext4 and worst: in the desktop land (where ZFS has little to offer compared to what it requires, and where it doesn't aim)
        Is more reasonable to compare it against something of similar kind, like btrfs.

        But on the areas of NAS, Storage providers, and other kinds of servers who could benefit of features like dedupe or transparent compression and/or you have some gear in there ... you win, in a manner that cannot be expressed

        Regards.

        Comment


        • #19
          Originally posted by uid313 View Post
          The premise of ZFS was never that it was blazing a blazing fast high-performance file system.

          The premise of ZFS is scalability, manageability, stability, data security, and features.

          ZFS would be great as a NAS, a media server (with big files such as movies), and file system for backup and archival.

          For the last use cases, at least, SnapRAID is at least as good of a solution as ZFS, plus there is nearly zero chance that you will lose your entire array since it doesn't stripe the data (it does something more like the old WHS Spaces or Unraid where they set aside disks to hold parity).

          Comment


          • #20
            Btrfs will be tested enough when Oracle itself will approve it for their database. Fedora is RH testing playground, so they don't care if you loose all your data.
            Last edited by Stellarwind; 09-23-2012, 07:13 AM.

            Comment

            Working...
            X