Announcement

Collapse
No announcement yet.

Benchmarks Of The Liquorix Linux 4.17 Kernel

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

  • Benchmarks Of The Liquorix Linux 4.17 Kernel

    Phoronix: Benchmarks Of The Liquorix Linux 4.17 Kernel

    It's been a while since last benchmarking the Liquorix kernel that is a modified version of the Linux kernel. Liquorix was recently updated against Linux 4.17 and a premium patron requested some fresh benchmark results...

    http://www.phoronix.com/scan.php?pag...ux-4.17-Kernel

  • #2
    I wish there were more latency/stability tests...

    (you missed OpenArena)

    Comment


    • #3
      can someone explain, why one of the Compile Bench tests yielded 100% more performance compared to the other kernels?

      Comment


      • #4
        Originally posted by jakubo View Post
        can someone explain, why one of the Compile Bench tests yielded 100% more performance compared to the other kernels?
        Completely wild guess here, but could be the 1000HZ scheduling in combination with a poorly supported IO device (masked DMA / polling) that is stuck in spinlocks (IOWAIT) a large percentage of the time. In that case high frequency kernel preemption would allow it to do meaningful compilation on the core that's slaved to that IO device, and that could double throughput in some circumstances on a low core count machine depending on the storage setup. It can also significantly harm performance in other circumstances. With the various Spectre mitigations in place already hurting IO performance, throwing in 1000HZ syncronous scheduling is going to harm rapid metadata updates like the Initial Create test, Postfix tests, and SQLite because they cause a lot of context switches.

        Comment


        • #5
          Originally posted by tildearrow View Post
          I wish there were more latency/stability tests...
          Yes it's very hard to profile this. The stated goal of the Liquorix configuration is "responsiveness at the cost of throughput and power usage." So the expectation is to sacrifice some performance on these kinds of benchmarks, while gaining something that is harder to quantify.

          Comment


          • #6
            These tests could not be more biased against the Liquorix kernel. They are simply ridiculous. First of all, you should test a weak cpu, not one of the best if not the best one simply can't struggle. Secondly, there should be no graphical benchmark. They won't be accurate anyway since graphic benchmarks almost always need developer fine tuning. You can bet nobody has fine tuned any of their graphic benchmarks for Liquorix; no-one. There is no a single benchmark in your test suite to measure responsiveness. Not one. Of course, responsiveness in relation to multi-tasking would be prime real estate for the Liquorix kernel. Overall, the article is an embarrassment to the expected standard @ Phoronix.

            Comment

            Working...
            X