Announcement

Collapse
No announcement yet.

Btrfs RAID 0/1/5/6/10 Five-Disk Benchmarks On Linux 4.1

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

  • Btrfs RAID 0/1/5/6/10 Five-Disk Benchmarks On Linux 4.1

    Phoronix: Btrfs RAID 0/1/5/6/10 Five-Disk Benchmarks On Linux 4.1

    Earlier this month I posted some Btrfs RAID 0/1 benchmarks on Linux 4.1 as a prelude to some larger Btrfs RAID benchmarks. Today the rest of those results are available with using five disks and testing Btrfs on this newest version of the Linux kernel while testing the RAID 0, 1, 5, 6, and 10 levels.

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Hi!

    Thanks for benchmarking Btrfs RAIDs for us!

    I am surprised by how badly RAID10 performs compared to RAID5 in the 5-disk setup. Quite unexpected for me. Is it me or should RAID10 be faster than any parity-based solution? Can you comment on this?

    Also would it be possible to included rebuild time for degraded arrays filled at at least 50% capacity?

    Thanks a lot! Keep'em benchmarks coming!

    V
    Last edited by mup4; 29 May 2015, 03:06 PM.

    Comment


    • #3
      Thanks for the benchmark!
      Looks like there is still some work to be done on RAID6. Performance is significantly worse than RAID5.

      Comment


      • #4
        Hi, thanks for the benchmark, but unfortunately for me, it's completely un-trustworthy. The reason is simple: see first graph of sequential read. You claim single-disk to have 41.3 MB/s speed and two disks in RAID0 (stripping) do have more than 2x speed of the single-disk. In fact it's nearly 4x so something you would expect from 4x drive in RAID0 setup. This is not possible engineering-wise IMHO. Either the single-disk drive was also dying or some other problem happen in your benchmarking (wrong setup?). See here for more reliable numbers -- got from FreeBSD/ZFS in this case, but those are IMHO numbers you can get from the technology realistically. https://calomel.org/zfs_raid_speed_capacity.html

        Comment


        • #5
          Something is definitely weird with first 2 charts -- all of the writes are faster than reads!

          Comment


          • #6
            Originally posted by mup4 View Post
            Hi!

            Thanks for benchmarking Btrfs RAIDs for us!

            I am surprised by how badly RAID10 performs compared to RAID5 in the 5-disk setup. Quite unexpected for me. Is it me or should RAID10 be faster than any parity-based solution? Can you comment on this?

            Also would it be possible to included rebuild time for degraded arrays filled at at least 50% capacity?

            Thanks a lot! Keep'em benchmarks coming!

            V
            This is an surprising result indeed. It almost seems like it could be a benchmark artefact. It would be interesting to hear from someone who is running BTRFS RAID5 in a real world setting.

            Comment


            • #7
              This being said, I do not want to discourage Phoronix from the significant effort to produce those benchmark. Can someone on the kernel mailing list comment on this? We are talking about unreleased code here. Thanks!

              Comment


              • #8
                There are other odd things.

                E.g.: Flexible IO Tester v2.1.13 read:
                - 5 disk raid0 is slower than 4 disc raid0
                - but 5 disc raid 5 is more than double the speed of 4 disc raid 5

                how large was the variance of the tests here? how often where the tests repeated?

                Comment


                • #9
                  Originally posted by tillschaefer View Post
                  how large was the variance of the tests here? how often where the tests repeated?
                  The variance/deviation is indicated on all of the graphs, as always. With Phoronix Test Suite the tests are done always at least three times but will dynamically do more if the variance is high, etc. It's all reproducible with PTS.
                  Michael Larabel
                  https://www.michaellarabel.com/

                  Comment


                  • #10
                    Originally posted by tillschaefer View Post
                    There are other odd things.

                    E.g.: Flexible IO Tester v2.1.13 read:
                    - 5 disk raid0 is slower than 4 disc raid0
                    - but 5 disc raid 5 is more than double the speed of 4 disc raid 5

                    how large was the variance of the tests here? how often where the tests repeated?
                    You'd have to run PTS to see it, but every benchmarks records deviation. It's all right there in front as the benchmarks run. It's pretty good stuff.

                    Comment

                    Working...
                    X