Results 1 to 10 of 24

Thread: ZFS File-System For Linux Is Still Around

Hybrid View

  1. #1
    Join Date
    Jan 2007
    Posts
    14,909

    Default ZFS File-System For Linux Is Still Around

    Phoronix: ZFS File-System For Linux Is Still Around

    While Btrfs, XFS, and EXT4 remain the far more popular choices when it comes to Linux file-systems, there still exists projects focused upon providing ZFS file-system support under Linux...

    http://www.phoronix.com/vr.php?view=MTA5NzQ

  2. #2
    Join Date
    Sep 2008
    Posts
    89

    Default

    However, with FreeBSD, OpenIndiana, and Solaris 11 around - do you really want to?

  3. #3
    Join Date
    Aug 2010
    Posts
    28

    Default ZFS On Linux Recognition

    I've asked myself why there's been little recognition from Phoronix to recognize the ZoL ( ZFS on Linux ) project over the last year or so. It's good to see a writeup again, and hopefully we'll see some benchmarks soon. I'd like to see ZoL compared to ZFS Fuse and ZFS under a BSD.

    A lot of work has been done on this project and the forums are very helpful. ZoL performs extremely well for myself and many others. Illumos ZFS bugs are frequently merged into the ZoL code base too.

    The latest version even provides the .zfs/snapshot folder where you can see all your snapshots without having to do any additional mounts--great stuff!

    It's very simple and quick to install from source code too. Checkout, compile, and install can be done within 10 minutes time for anyone who has ever run a ./configure script and compiled source code.

    It's unfortunate with the licensing issue that it can't be stock included with the kernel, but all the source code is available, and easy to get up an running on any system. I've lived with binary NVIDIA blobs for years now, so I don't have an issue having to do an occasional git pull and recompile/install to pull in the latest code from ZoL .

    While there are many features and knobs to tune and control ZFS, the only tunable I need to make it work great for me is to limit the amount of memory used for the ARC . A modprobe.d file option:

    options zfs zfs_arc_max=1610612736

    works for me to limit the amount of memory of the cache.

    I've been running ZoL for a while now on a 5 disk 4TB ( 2 mirrored pairs, 1 hot spare, and 4 old 1GB usb flash cache devices ) array using several compressed and uncompressed filesystems. I perform all my nightly backups there from several computers, run a postgres tablespace with compression on it, and run a few KVM images off of it, and serve up video files for streaming to several devices. I've not run into any issues.

  4. #4
    Join Date
    Aug 2010
    Posts
    59

    Default

    ZfsonLinux requires your kernel to be non SMP.

  5. #5
    Join Date
    Aug 2010
    Posts
    28

    Default

    SMP works just fine. The restriction is on the preemption model: You can't use CONFIG_PREEMPT ( low latency desktop ) , you need either CONFIG_PREEMPT_VOLUNTARY or CONFIG_PREEMPT_NONE .

    Linux xxxxxx 3.3.4+ #8 SMP Fri Apr 27 15:30:00 MDT 2012 x86_64 x86_64 x86_64 GNU/Linux

  6. #6
    Join Date
    Feb 2008
    Posts
    210

    Default

    If ZFS wants to survive outside Solaris and their Open Source forks, a dual licensing must be made in some form and included into the vanilla kernel.

    Other than that, this is like the fanatics wanting to use Reiser4 into Linux 3.x. It's just a pipe dream that's going nowhere.

    Please be more realistic, even HAMMER2 is a more viable option once Matt Dillon agrees on some kind of interoperatibility between non-BSD UNIX-like systems.

    But what about Btrfs?

    What about distributed file systems? CRFS, FhGFS, Tahoe-LAFS, Ceph, Lustre, MooseFS, GFS2, OCFS, OneFS, XtreemFS, GlusterFS, HAMMER2 (seems it will have those features), pNFS, AFS.

    What I mean is not just a new, scalable and proper and efficient network filesystem, but also RAID-like capabilities. That could make commodity hardware to get into cheap RAID solutions to avoid data losing.



    So what about an article about that instead beating a dead horse like ZFS? Despite all the hype in the past, it reduces interoperability between al UNIXes (and non-UNIXes).

  7. #7
    Join Date
    Aug 2010
    Posts
    59

    Default

    Quote Originally Posted by mgmartin View Post
    SMP works just fine. The restriction is on the preemption model: You can't use CONFIG_PREEMPT ( low latency desktop ) , you need either CONFIG_PREEMPT_VOLUNTARY or CONFIG_PREEMPT_NONE .

    Linux xxxxxx 3.3.4+ #8 SMP Fri Apr 27 15:30:00 MDT 2012 x86_64 x86_64 x86_64 GNU/Linux
    Sorry that's what I meant. Tried it and vary bad for a desktop multitasking situation.

  8. #8
    Join Date
    Dec 2007
    Posts
    146

    Default

    Quote Originally Posted by joffe View Post
    However, with FreeBSD, OpenIndiana, and Solaris 11 around - do you really want to?
    I've got a 5.something Terrabyte raidz array in my HTPC. I've had it running well over a year now with few issues. The data integrety has been perfect, however, there is an annoying bug that causes the system to hang from time to time if performing small write operations on one of the files over Samba.

    In theory, for reliable software raid 5, raidz can't be beat. I think practice is still catching up to theory, but I hope they keep working on it.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •