Announcement

Collapse
No announcement yet.

QEMU 1.0 Is Coming Quite Soon

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

  • QEMU 1.0 Is Coming Quite Soon

    Phoronix: QEMU 1.0 Is Coming Quite Soon

    Version 1.0 of QEMU will be released next month in time for the holidays with several interesting advancements. QEMU is the popular open-source machine emulator and virtualizer that also plays a role in the Linux KVM virtualization stack...

    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
    kvm-init?

    Very cool. I just got an idea and some fast googling did not come up with an obvious answer...

    The idea was basically: is it possible to make an extremely minimal kvm/qemu host, where basically a guest OS is booted directly from the kernel by a custom init (modifiable from the guest OS) without the overhead of a userland from the host OS?.

    Comment


    • #3
      Yes, you can launch qemu from an init script (note: not init.d, /sbin/init) directly. The guest-modifiable part is possible too, with some network sharing etc.

      What do you have in mind?

      Comment


      • #4
        Originally posted by curaga View Post
        Yes, you can launch qemu from an init script (note: not init.d, /sbin/init) directly. The guest-modifiable part is possible too, with some network sharing etc.

        What do you have in mind?
        Primarily I was thinking about an extremely minimal linux host as a "hardware padding layer" for OSes with less developed HW support, and still keep as much resources free as possible for the guest OS making it an almost-bare-metal install.

        (and perhaps even more naively, this minimal host could perhaps be compiled and mantained within the guest OS, for example a BSD variant using GCC or Clang or some other OS able to run the required components to compile Linux and whatever more that would be required for a minimal host. As I said, I have not thought it through yet but it felt like a cool idea.)

        Comment


        • #5
          Originally posted by staalmannen View Post
          Very cool. I just got an idea and some fast googling did not come up with an obvious answer...

          The idea was basically: is it possible to make an extremely minimal kvm/qemu host, where basically a guest OS is booted directly from the kernel by a custom init (modifiable from the guest OS) without the overhead of a userland from the host OS?.
          Funny you should ask. A fedora guy (daniel berrange http://berrange.com/posts/2011/09/22...rnel-shutdown/) was blogging about this very thing. He is working on creating qemu/kvm sandboxes for certain applications (say a browser) that literally loads a new initrd but shares the root directory (read-only) and provides overlays for some other directories (proc/sys at a min). For you geeks out there, in order to passthrough these filesystems to qemu he uses the p9p virtio! At the point of that post, he had it booting in about 2.8s (with 2s due to seabios/qemu itself). Because of this work, he found this weird behavior in MD driver where it ALWAYS waited for at least 1s to see if an MD device was present! Once he got rid of that he managed to make it shutdown very quickly as well.

          Best/Liam

          Comment


          • #6
            Originally posted by liam View Post
            Funny you should ask. A fedora guy (daniel berrange http://berrange.com/posts/2011/09/22...rnel-shutdown/) was blogging about this very thing. He is working on creating qemu/kvm sandboxes for certain applications (say a browser) that literally loads a new initrd but shares the root directory (read-only) and provides overlays for some other directories (proc/sys at a min). For you geeks out there, in order to passthrough these filesystems to qemu he uses the p9p virtio! At the point of that post, he had it booting in about 2.8s (with 2s due to seabios/qemu itself). Because of this work, he found this weird behavior in MD driver where it ALWAYS waited for at least 1s to see if an MD device was present! Once he got rid of that he managed to make it shutdown very quickly as well.

            Best/Liam
            Awesome!

            And for the p9p touch, doubly so! (I am sort of a wannabe Plan9 user playing in 9vx + dual boot 9front at the moment, but still pretty much a noob, was planning to try to get php to compile under APE to try to run pts on it, but that has so far failed despite being a pure C program).

            Comment


            • #7
              Originally posted by staalmannen View Post
              Awesome!

              And for the p9p touch, doubly so! (I am sort of a wannabe Plan9 user playing in 9vx + dual boot 9front at the moment, but still pretty much a noob, was planning to try to get php to compile under APE to try to run pts on it, but that has so far failed despite being a pure C program).
              I've never actually played with plan9 directly, but we've all been able to take advantage of some of its ideas in linux (further cementing linux's reputation as a mutt os). As we can see, p9's influence continues. This further solidifies Richie's reputation.

              Comment

              Working...
              X