Announcement

Collapse
No announcement yet.

Btrfs RAID vs. Linux Software RAID Benchmarks On Linux 4.7

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

  • Btrfs RAID vs. Linux Software RAID Benchmarks On Linux 4.7

    Phoronix: Btrfs RAID vs. Linux Software RAID Benchmarks On Linux 4.7

    Earlier this month I carried out some 4-disk Btrfs RAID benchmarks using four SATA 3.0 SSDs. Those tests were done using the Btrfs built-in RAID capabilities while today are some comparison tests against those numbers when using the Linux Software RAID setup via mdadm.

    http://www.phoronix.com/vr.php?view=23435

  • #2
    for me the BTRFS results are impressive... in theory BTRFS should be much slower because of the "Copy on Write" concept. but the more harddrives the faster it becomes really impressive.
    Phantom circuit Sequence Reducer Dyslexia

    Comment


    • #3
      Originally posted by Qaridarium View Post
      for me the BTRFS results are impressive... in theory BTRFS should be much slower because of the "Copy on Write" concept. but the more harddrives the faster it becomes really impressive.
      The RAID-1 concepts are different. In BTRFS, RAID-1 is filesystem RAID-1, so there are always only two copies of the data no matter how many drives. MD RAID-1 mirrors the data to each drive (four copies in a four-drive RAID-1). That's why BTRFS gets faster; only two writes per written data block vs. N-writes.

      Comment


      • #4
        Originally posted by Qaridarium View Post
        for me the BTRFS results are impressive... in theory BTRFS should be much slower because of the "Copy on Write" concept. but the more harddrives the faster it becomes really impressive.
        I assume that btrfs was the filesystem used on top of the mdadm setup as well (did not bother to read the benchmark). I have no idea for what duration the tests where run, but from my experience btrfs performance drops once the disks becomes filled up and therefore fragmented. I am a fan of BTRFS and I am currently running tests on it in raid10 and on my system I will have a performance of about 750mb/s but have already run into issues with free space when trying to rebalance to raid1. While BTRFS is promising in concept even the basics are far from perfect.

        Comment


        • #5
          Originally posted by macemoneta View Post

          The RAID-1 concepts are different. In BTRFS, RAID-1 is filesystem RAID-1, so there are always only two copies of the data no matter how many drives. MD RAID-1 mirrors the data to each drive (four copies in a four-drive RAID-1). That's why BTRFS gets faster; only two writes per written data block vs. N-writes.
          That actually sounds like it's using a RAID-10 past the first two discs if it's performance scales linearly (excluding the usual multi-drive overhead drop-off) under RAID-1. Not an ideal situation if you've planned on a 4-disc array because your data matters that much. Thanks for the pointer as it's worth investigating.

          Comment


          • #6
            Originally posted by stiiixy View Post
            Not an ideal situation if you've planned on a 4-disc array because your data matters that much.


            RAID != backup

            If your data matters at all you'll be needing backup.

            Comment


            • #7
              Originally posted by Dick Palmer View Post



              RAID != backup

              If your data matters at all you'll be needing backup.
              I fully agree with you. An array of disks is meant so that machine stays functioning on drive failure. Data backup is something totally different. It shocks me how many people implement RAID as their only means of data protection.

              Comment


              • #8
                Does btrfs implement paralel read for RAID1? AFAIK single process (thread?) will read only from single disk currently.

                Comment


                • #9
                  Originally posted by duby229 View Post

                  I fully agree with you. An array of disks is meant so that machine stays functioning on drive failure. Data backup is something totally different. It shocks me how many people implement RAID as their only means of data protection.
                  You're both missing the point. This isn't about best practices and about BTRFS having some quirks that are unfortunate and scare people away who dont want to use it as the primary storage/daily driver. I dont want to be the guy who has to explain why you constantly need to restore from back ups ($$ being burned on time, potential corrupt data issues, blahblah) nor would I want to restore from backups if it's simply a quirky FS if there are other systems around that do it better, faster and more reliably.

                  Comment


                  • #10
                    MD RAID-10 results seems to be way off and just can't be true that MD RAID10 is slower then MD RAID1. I guess that there is some big configuration problems, miss alignment of the partition or something else I guess.
                    @Michael - If you have time you can try to erase the 4x SSD's recreate the MD-RAID10 array and re-run the tests, I will be very curious what will be the results.

                    Comment

                    Working...
                    X