Announcement

Collapse
No announcement yet.

BFQ Scheduler Will Try To Go Mainline In The Linux Kernel

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

  • BFQ Scheduler Will Try To Go Mainline In The Linux Kernel

    Phoronix: BFQ Scheduler Will Try To Go Mainline In The Linux Kernel

    The developers behind the BFQ I/O scheduler are preparing patches to try to mainline the scheduler within the upstream Linux kernel...

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

  • #2
    I find this scheduler meaningless after I started using xfs on my system partitions and when kernel 3.13 brought the io queuing layer.

    Comment


    • #3
      Originally posted by roberth View Post
      I find this scheduler meaningless after I started using xfs on my system partitions and when kernel 3.13 brought the io queuing layer.
      Maybe you should check the benchmarks on 3.13 http://algo.ing.unimo.it/people/paol...ed/results.php

      Comment


      • #4
        Originally posted by AnAkIn View Post
        Maybe you should check the benchmarks on 3.13 http://algo.ing.unimo.it/people/paol...ed/results.php
        I use an ssd for system partitions, and the benchmark shows no gain with bfq and ssds AFAICT.

        Comment


        • #5
          Originally posted by roberth View Post
          I find this scheduler meaningless after I started using xfs on my system partitions and when kernel 3.13 brought the io queuing layer.
          What's this new io queuing layer? Link please?
          Thanks!

          Comment


          • #6
            Originally posted by kokoko3k View Post
            What's this new io queuing layer? Link please?
            Thanks!
            *Multi-Queue Block Layer

            http://www.phoronix.com/scan.php?pag...tem&px=MTUxNDQ

            Comment


            • #7
              Thank you.

              Comment


              • #8
                Originally posted by roberth View Post
                I use an ssd for system partitions, and the benchmark shows no gain with bfq and ssds AFAICT.
                Look again, it does for startup times.

                Comment


                • #9
                  Finally a usable, low-latency I/O scheduler for linux. I hope the kernel developers get this merged ASAP.

                  Comment


                  • #10
                    Originally posted by AnAkIn View Post
                    Look again, it does for startup times.
                    What surprises me is that from my experience BFQ makes the boot up take longer. I tested this on my system with SSD and checked the startup time with systemd-analyze with and without "elevator=bfq" as kernel line parameter. Takes a good second longer to boot.

                    Comment


                    • #11
                      Originally posted by AnAkIn View Post
                      Maybe you should check the benchmarks on 3.13 http://algo.ing.unimo.it/people/paol...ed/results.php
                      Any independent benchmarks that show BFQ to be better than whats already in the kernel?

                      Originally posted by jwilliams View Post
                      Finally a usable, low-latency I/O scheduler for linux. I hope the kernel developers get this merged ASAP.
                      Wheres the proof that it is Useable, low-latency or faster? I tried it out a few weeks ago and it was slower than Deadline and CFQ, with higher latency. BFQ is nothing more than snake oil.

                      Comment


                      • #12
                        Originally posted by Rallos Zek View Post
                        Wheres the proof that it is Useable, low-latency or faster?
                        Works for me.

                        Comment


                        • #13
                          Originally posted by jwilliams View Post
                          Works for me.
                          And for me.

                          I don't have any SSDs or clever RAID setups, but application load times with disk-intensive stuff in the background (mostly compiling) are 50% or so better on my 7200rpm spinning rust.

                          Comment


                          • #14
                            Originally posted by FLHerne View Post
                            And for me.

                            I don't have any SSDs or clever RAID setups, but application load times with disk-intensive stuff in the background (mostly compiling) are 50% or so better on my 7200rpm spinning rust.
                            When my HDDs are under heavy load, it makes a lot more than 50% difference for me. Under heavy background I/O load, sometimes it can take 1 or 2 minutes for a small program to start with CFQ, but with BFQ, just the normal few seconds.

                            You can see the same thing in some of the tests on the BFQ website. Notice the "X" marks where several other I/O schedulers fail to complete the I/O within 60 seconds.

                            Before I found BFQ, I could hardly believe that the linux I/O schedulers were so bad that heavy I/O could virtually freeze interactive tasks on my computer. But it was the case for both cfq and deadline. Hence, my comment that finally the (mainline) linux kernel looks like it will have a decent I/O scheduler (and I will not have to use customize pf kernel's any more).

                            Comment


                            • #15
                              Originally posted by FLHerne View Post
                              And for me.

                              I don't have any SSDs or clever RAID setups, but application load times with disk-intensive stuff in the background (mostly compiling) are 50% or so better on my 7200rpm spinning rust.
                              ++

                              and for me

                              Comment

                              Working...
                              X