Announcement

Collapse
No announcement yet.

Patches Published Again For Replacing Linux Kernel's CFQ With BFQ

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

  • Patches Published Again For Replacing Linux Kernel's CFQ With BFQ

    Phoronix: Patches Published Again For Replacing Linux Kernel's CFQ With BFQ

    A set of 22 patches were published this week that seek to replace the Linux kernel's default I/O scheduler CFQ (Completely Fair Queueing) with BFQ, the Budget Fair Queueing...

    http://www.phoronix.com/scan.php?pag...Patches-Posted

  • #2
    Last time I tried BFQ with the linux-ck kernel on Arch Linux on my SSD the boot times were significantly worse (2-3 seconds) compared to CFQ. Right now I'm using noop.

    Comment


    • #3
      To make a smooth transition, this patchset first brings CFQ back to its state at the time when BFQ was forked from CFQ. Basically, this reduces CFQ to its engine, by removing every heuristic and improvement that has nothing to do with any heuristic or improvement in BFQ, and every heuristic and improvement whose goal is achieved in a different way in BFQ. Then, the second part of the patchset starts by replacing CFQ's engine with BFQ's engine, and goes on by adding current BFQ improvements and extra heuristics."
      when confronted with :

      On the BFQ site their results show BFQ performing significantly better than CFQ, Deadline, and Noop I/O schedulers for various storage mediums.
      makes me suspicious about how exactly the benchmarks were performed, as many youtube demos did switch schedulers on the fly. If the dumbed down cfq was used as comparison, no wonder bfq appears to be so great.

      I didn't poke through the code yet, though. Just running on paranoia atm.

      Comment


      • #4
        Originally posted by blackout23 View Post
        Last time I tried BFQ with the linux-ck kernel on Arch Linux on my SSD the boot times were significantly worse (2-3 seconds) compared to CFQ. Right now I'm using noop.
        Was this with an older version of BFQ? I remember such issue being fixed at some point.

        Comment


        • #5
          isnt this a benchmark site...? o.O just saying :P

          Comment


          • #6
            Likelyhood of adding more code to improve speed if I/O request (random or not) capacity exceeds I/O request need in typical application usage = close to 0.
            This is not going to change as we move from ACHI-SATA/SSD to PCIE/NVMe SSD in the near term.
            Rotational media with lousy I/O request capacity is another subject altogheter.

            Comment


            • #7
              Originally posted by jakubo View Post
              isnt this a benchmark site...? o.O just saying :P
              I remember the distros benchmarks where Manjaro used to be slower, and one of the hypothesis was that it were the only one running with BFQ instead of CFQ

              Comment


              • #8
                I use BFQ. With CFQ on Ubuntu's kernel, videos dont play smooth and freeze at times when file transfer from an external HDD to the one where movies are stored is in progress.(HDD runs on a SATA v2 bus and gives max 30MBps ). With BFQ, everything runs butter smooth. BFQ was the reason i was forced to recompile kernel for the very first time.

                I would like to see it get mainlined. Wont have to follow the patchset. Also how would this affect blk_mq ?

                Comment


                • #9
                  I honestly don't see what the whole deal with BFQ is. It seems to work for some people, judging by the comments and what I am reading, but it has never worked for me.

                  I tried switching to it many times before, but every time I noticed drastically worse I/O performance. My system would freeze up and lock up under heavy load, etc. And I am a heavy I/O user. I have a 6-hdd btrfs raid array on my desktop and 3 ssds (one nvme and two sata), and I often do various I/O heavy workloads on many of them simultaneously. Every time I tried BFQ, I switched back to CFQ a couple of days later because the performance with BFQ was noticeably worse. The whole system felt sluggish.

                  Comment


                  • #10
                    Just last week I switched on blk-mq. I'm not entirely sure what it is or if it replaces or works as a layer above/below CFS/BFQ/noop/deadline, but my subjective experience so far is a more responsive system and shorter loading times when using an SSD, compared to using deadline without blk-mq.
                    Last edited by Beherit; 04 February 2016, 12:29 PM. Reason: Forgot to add "..compared to using deadling"

                    Comment

                    Working...
                    X