Announcement

Collapse
No announcement yet.

Multigen LRU Patches Updated For Addressing Linux's Expensive Page Reclaimation

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

  • Multigen LRU Patches Updated For Addressing Linux's Expensive Page Reclaimation

    Phoronix: Multigen LRU Patches Updated For Addressing Linux's Expensive Page Reclaimation

    One of the interesting performance-related kernel patch series to come about so far this year has been Google's multi-generational LRU framework that is promising to offer much better performance in addressing the kernel's expensive page reclaimation handling...

    https://www.phoronix.com/scan.php?pa...U-Framework-V3

  • #2
    This is awesome! Linux very bad low memory handling have been by biggest problem since starting to daily drive Linux in 2006. Even with 32GB RAM I eventually end up with a broken system than I hopefully can resurrect by manually triggering the OOM killer with SysRq + f but more often than not just have to force reboot from.

    Comment


    • #3
      I am testing it out right now, with X=7 and N=2 for a gaming desktop system. The previous patchset already made it into the Zen and Xanmod Kernels. The author is also very responsive and if some of you could share their success stories with him on LKML, he'd be delighted for sure.

      Comment


      • #4
        That's very nice and all, but I don't have any RAM pressure issues.
        Touted as "less heavy on the CPU".
        Does it mean that it performs better even when not under memory pressure?
        "Normal usage" performance stats? Will I notice any difference?

        Better? Worse? Same?

        Comment


        • #5
          Originally posted by milkylainen View Post
          That's very nice and all, but I don't have any RAM pressure issues.
          Touted as "less heavy on the CPU".
          Does it mean that it performs better even when not under memory pressure?
          "Normal usage" performance stats? Will I notice any difference?

          Better? Worse? Same?
          From the discussion I read on the v2 it seems in fact to work better even when there's no memory pressure, because there's less contention on some lock. At least that's how I recall the discussion as someone with no knowledge on the MM subsystem internals.

          Comment


          • #6
            My success story: I have Archlinux with 8G RAM + zswap + swap […]
            Yaay, so cool, this was mine story ^^ I was hoping to give feedback just to make more motivation to people working on that, so the patch series wouldn't linger for another year as it sometimes happens. Though, at this point it is apparently a popular patchset as it's included in linux-zen, which also means there's no need for me to build my own kernel anymore.

            Btw, yeah, I'm still using this patchset as well as built it for my gf on Fedora 33, and our experience is amazing. Especially relevant to her as her laptop only has 4G RAM.
            Last edited by Hi-Angel; 20 May 2021, 09:23 AM.

            Comment


            • #7
              Originally posted by Hi-Angel View Post
              Especially relevant to her as her laptop only has 4G RAM.
              I remember the days when 4Gb was literally more than my system could handle. Now it's just 2/3 of Chrome.

              Comment


              • #8
                Ooooh, this sounds very useful for my laptop! Stuck with 8 GB of RAM there, and doing the same swap + zswap approach. Hopefully this can improve the situation considerably.

                Comment


                • #9
                  count me in:

                  8 GiB RAM soldered onboard
                  + 8 GiB RAM module
                  + 8 GiB zramswap (50% of total RAM-size)
                  + 64 GiB SSD-swap for hibernation and leeway

                  Comment


                  • #10
                    Also in a RAM constrained machine
                    8 GB RAM, 1,5 GB ZRAM (LZ4), 9 GB HDD swap

                    I noticed that the system sometimes had trashing like behaviour with high CPU usage whenever ZRAM started to be used, but I thought it was normal. The slowdowns tend to happen whenever I run Gradle stuff with Android Studio and web browsers on when Vivado tries to eat away all memory trying to optimise stuff.

                    Comment

                    Working...
                    X