Announcement

Collapse
No announcement yet.

Linux 4.12 I/O Scheduler Tests With A HDD & SSD

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

  • Linux 4.12 I/O Scheduler Tests With A HDD & SSD

    Phoronix: Linux 4.12 I/O Scheduler Tests With A HDD & SSD

    Earlier in the Linux 4.12 cycle I delivered a number of I/O scheduler benchmarks from a solid-state drive while for this article are some fresh I/O scheduler tests using a slower SSD as well as a conventional HDD. Schedulers tested were CFQ, Noop, deadline, MQ None, MQ Kyber, MQ BFQ, and MQ Deadline.

    http://www.phoronix.com/vr.php?view=24887

  • #2
    Looks like BFQ is the best for those who dont want to wait for disk operations. e.g. desktop users.

    Comment


    • #3
      Huh. I knew BFQ was latency-oriented but I did not expect it to be that bad at throughput.

      Perhaps something about multi-queue mode not being handled properly? I would really like an explanation.

      Comment


      • #4
        Where is the CFQ result on the last graph (the postgres benchmark) ?

        Comment


        • #5
          Originally posted by Shnatsel View Post
          Huh. I knew BFQ was latency-oriented but I did not expect it to be that bad at throughput.

          Perhaps something about multi-queue mode not being handled properly? I would really like an explanation.
          If you are constantly context switching then you will suffer in terms of total work throughput - but you may appease some managers (wait, are we still talking about schedulers here?)

          Comment


          • #6
            Originally posted by Shnatsel View Post
            Huh. I knew BFQ was latency-oriented but I did not expect it to be that bad at throughput.

            Perhaps something about multi-queue mode not being handled properly? I would really like an explanation.
            Or there was some other application(s) which BFQ gave higher priority during the tests.

            This sounds kinda strange though, in my tests I didn't get this massive performance hits. Although I did much more mundane things, without the test suite.

            Comment


            • #7
              also blk-mq seems to have a bug that hurts anything multiqueue

              Comment


              • #8
                Michael This probably comes often when people test BFQ, but can you try running the tests with the low_latency parameter disabled?

                Comment


                • #9
                  What's up with IOzone graph? Since when is SSD slower than HDD?

                  Comment


                  • #10
                    Originally posted by Fry-kun View Post
                    What's up with IOzone graph? Since when is SSD slower than HDD?
                    didn't notice, WTF IMPOSIBURU!

                    Comment

                    Working...
                    X