Announcement

Collapse
No announcement yet.

BFQ Gets Another Notable Responsiveness Fix

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

  • BFQ Gets Another Notable Responsiveness Fix

    Phoronix: BFQ Gets Another Notable Responsiveness Fix

    In addition to the BFQ improvements already staged for Linux 4.14, it looks like another fix will be on the way...

    http://www.phoronix.com/scan.php?pag...eness-Test-Fix

  • #2
    Nice, those wishing to see BFQ in Ubuntu's kernel, vote for https://bugs.launchpad.net/bugs/1715758 please

    Comment


    • #3
      Please make those tests on HDD too, where I/O schedulers have higher impact.

      Comment


      • #4
        Michael, could you clarify if the graph in this article is BFQ before the fix or BFQ after the fix?

        Comment


        • #5
          Originally posted by Zan Lynx View Post
          Michael, could you clarify if the graph in this article is BFQ before the fix or BFQ after the fix?
          It is a copy of the graph without the fix (I checked).

          But I agree, a simple sentence explaining this would be most welcome -- something like: "The graph below shows the results without the fix."

          Comment


          • #6
            Proper link to initial submission: https://github.com/Algodev-github/bf...97202856e1a7d6

            Updated link to ported commit in my tree: https://github.com/pfactum/pf-kernel...5aa9e43788b723

            Comment


            • #7
              The updates mentioned for BFQ are for the SQ scheduler or the MQ?(or both?)

              Comment


              • #8
                Originally posted by polarathene View Post
                The updates mentioned for BFQ are for the SQ scheduler or the MQ?(or both?)
                Both. Anyway, this fix is a work in progress, and we have already found a regression. Working on it right now ...

                Comment


                • #9
                  Originally posted by paolo View Post

                  Both. Anyway, this fix is a work in progress, and we have already found a regression. Working on it right now ...
                  Oh that's great to hear So in general both single and multi-queue BFQ schedulers will have parity? I've yet to switch to blk-mq, not sure if there is an advantage for a single SSD(haven't looked into it).

                  Thank you for the excellent I/O scheduler for desktop users It's a great experience!

                  Comment


                  • #10
                    Originally posted by polarathene View Post

                    Oh that's great to hear So in general both single and multi-queue BFQ schedulers will have parity? I've yet to switch to blk-mq, not sure if there is an advantage for a single SSD(haven't looked into it).
                    I believe that the advantages are in doing IO from multiple CPU cores. By having multiple queues, cores can submit IO requests without stacking behind another core. And in really high performance networked or NVMe applications, multiple queues keep cores from stepping on each other's memory and cache lines.

                    From what I've read a single SSD can benefit from up to 10 queues, but usually its limited to one queue per core, not including hyperthread pairs. So a 4c/8t CPU would get 4 IO queues, as I understand it.

                    Comment

                    Working...
                    X