Announcement

Collapse
No announcement yet.

MD RAID Optimizations, Btrfs Fixes For Linux 4.11

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

  • MD RAID Optimizations, Btrfs Fixes For Linux 4.11

    Phoronix: MD RAID Optimizations, Btrfs Fixes For Linux 4.11

    The MD pull request was submitted on Friday for the Linux 4.11 kernel as were the Btrfs file-system changes...

    http://www.phoronix.com/scan.php?pag...rfs-Linux-4.11

  • #2
    I'm not mad about no new features in btrfs.
    In fact I'd be happy to see no additions and only bugfixes vor the next 3 years and get a btrfs with better performance and which I would even somewhat trust my data.
    Too often I had to scrap my btrfs tries because of well known bugs that just happen after a while, especially if one uses the unique features (like subvolumes).

    Comment


    • #3
      We are using btrfs in a 4 HD disk RAID10 configuration without significant problems for multiple years now. But we would really like to just throw an additional 2 SSD's at it and take advantage of hot data tracking, as being promised for the same multiple years on https://btrfs.wiki.kernel.org/index.php/Main_Page. Yeah, I know there are other ways to use ssd's as a cache for hdd's (bcache comes to mind), but not with the expected ease that btrfs would have of just adding an additional device (btrfs device add /dev/sdx /).

      So, I agree, nothing new, just fixing. And this.

      Comment


      • #4
        The one thing I'm reaaally waiting for is the snapshot-aware defrag. Although, I think plans for that were scrapped =\

        Comment


        • #5
          The one thing I'm reaaally waiting for is the RAID 5 fixes from Qu Wenruo. But as far as I've observed nobody seems to have pulled this into a kernel patchset yet :-(

          Comment


          • #6
            Originally posted by renkin View Post
            The one thing I'm reaaally waiting for is the snapshot-aware defrag. Although, I think plans for that were scrapped =\
            Me too. Will we ever get it? I think not...
            ## VGA ##
            AMD: X1950XTX, HD3870, HD5870
            Intel: GMA45, HD3000 (Core i5 2500K)

            Comment


            • #7
              I do use BTRFS without any fancy features e.g. no subvolumes (except the default one), no compression, no defrag etc... I have 5 disks set up as RAID1 for both data and metadata and it works great and have even saved me for a corruption or two (according to the logs).

              There are a few things that BTRFS in my oppinion should get straight before moving ahead with fancy features and that is:

              1. Use the device ID that BTRFS itself assign to the device for identifying disks.
              This is important since if you have a multi device setup with /dev/sdx, /dev/sdy and /dev/sdz for example and let's pretende that /dev/sdy disappears and reappears again as /dev/sdt. What will happen? Well, BTRFS will happily continue to try to write to /dev/sdy since it does not recognize that device id 2 (which it still recognize the disk as) have moved around a bit.

              2. Auto-reject disks based on number of faults. So far there is no way to tell BTRFS to discard a device if there is a high number of read/write/checksum/corruption errors. If anything you need to write something that polls the output of btrfs de st /mountpoint which again is not always as easy since BTRFS sometimes do NOT update the statistics on auto-repair unless you manually run a scrub! (I was told this yesterday on IRC).

              3. Reliable RAID 5/6 that allows you to save some space on larger disk arrays.

              4. Additional parity devices (The author of SnapRAID many years ago submitted patches that would allow for up to 6x parity devices! (WOW!). These where never merged which sucks bigtime!)

              5. Per subvolume mount options (with hopefully configurable "RAID" levels)

              6. All the other fancy stuff

              I think that for most people the reliability features is the biggest selling point. Sure you have snapshots so you can do rollbacks, compression, deduplication etc etc.... but those are utterly useless if you can't trust your data anyway. While BTRFS does work fine if you know what you are doing (which most people don't know) there are several things that I feel is not getting the attention it should.

              Comment

              Working...
              X