Announcement

Collapse
No announcement yet.

KVM Smokes VirtualBox On Initial AMD EPYC Linux Tests

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

  • KVM Smokes VirtualBox On Initial AMD EPYC Linux Tests

    Phoronix: KVM Smokes VirtualBox On Initial AMD EPYC Linux Tests

    I've been working on some AMD EPYC virtualization tests on and off the past few weeks. For your viewing before ending out the new year are some initial VirtualBox vs. Linux KVM benchmarks for seeing how the guest VM performance compares.

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    It would be very interesting to also include result numbers for the same benchmarks run directly on the physical host hardware (so outside of virtual machines).

    Comment


    • #3
      Originally posted by dwagner View Post
      It would be very interesting to also include result numbers for the same benchmarks run directly on the physical host hardware (so outside of virtual machines).
      Yes I plan to do that when running 64 vCPUs.... Basically when it matches the CPU configuration. Didn't think in the real world many people would boot EPYC with maxncpus=16 or so, so didn't do it for this round of testing and since wasn't sure how setting the maxncpus would map nicely for EPYC's design or not, etc.
      Michael Larabel
      https://www.michaellarabel.com/

      Comment


      • #4
        I'd also like to see the opensource xen performance thrown into the mix as well.

        Comment


        • #5
          Awesome benchmarks, thanks. Is it possible to also test PCI passthrough with a GPU for both Xen and KVM? A comparison to native would also be appreciated to compare the overheads.

          There should be no problems with AMD cards, but nVidia's driver (at lest on Windows) will detect a hypervisor and refuse to load. A workaround for that is to tell the hypervisor to either lie about the computer manufacturer or to not expose VM-specific interfaces to the guest at all (https://wiki.archlinux.org/index.php...to_Windows_VMs).

          Comment


          • #6
            Any info on KVM vs Virtualbox on Intel?

            Comment


            • #7
              Originally posted by ravenet View Post
              I'd also like to see the opensource xen performance thrown into the mix as well.
              libxenlight was giving me problems on Ubuntu 17.10 and didn't have time to dig deeper into it, will in the new year.
              Michael Larabel
              https://www.michaellarabel.com/

              Comment


              • #8
                Originally posted by kmand View Post
                Any info on KVM vs Virtualbox on Intel?
                WIll do some fresh Intel tests probably in a few weeks will find the time.
                Michael Larabel
                https://www.michaellarabel.com/

                Comment


                • #9
                  Is it possible some of the results have to do with the CPU extensions exposed by VM to the guest? Could you compare /proc/cpuinfo across bare metal/virtualbox/kvm? I remember AVX not being enabled by default on virtualbox, but that might have changed some time ago.

                  Comment


                  • #10
                    Originally posted by phoronix View Post
                    before ending out the new year
                    You probably mean "the year".

                    Comment

                    Working...
                    X