KVM84 in the starting blocks

Remember my post from March? Where, we're nearly there. Dustin announced a few days ago that he was expecting to push kvm84 in -updates next week. I've been beta-testing and chasing bugs on this for some time now, and I am pretty happy with this backport, and all the goodness it brings. So, a short list of features/bugfixes that I've noticed so far:

• Disk speed seems a lot better:

  yhamon@yhamon-dev:~$ time cp SunStudio12ml-solaris-sparc-200709-pkg.tar.bz2  tmp
  real	0m12.528s
  
  VS
  
  yhamon@mirror:~$ time cp SunStudio12ml-solaris-sparc-200709-pkg.tar.bz2 tmp
  real	0m20.159s
  

  yhamon-dev being a VM running on a kvm84 host, and the other one on a standard kvm-62. Both hosts are similar in specs and similarly loaded. I wonder what could trigger such a significant change (cache?), but I ran this cp several times, an every time kvm84 would be significantly faster...

• ACPI2 for windows guests: among other features, it means that windows guests will now be able to reboot "themselves". Until now, when triggering a reboot from a windows guest, it would just shut it down. Now that works fine, too.
• Proper SMP support. In KVM-62, SMP support was quite broken, it would use a lot of CPU on the host - the network would also regularly crash with SMP guests, leaving them without connectivity. Now this seems to work correctly; I've been running SMP windows and Linux guests for a while, and it seems quite stable.

Speaking of Windows guests (believe me, I'd be happier without): I think I found a better way to install those. Instead of using virt-install like I documented a while ago , it is much easier to directly manually create the QCOW2 file:

qemu-img create -f qcow2 windows.qcow2 12G

And to create manually the XML definition file (copy it from a template - don't forget to change UUID and MAC address), and boot it directly with virsh. Here is an example of a libvirt XML file that I use.

This works fine; I had many issues before with virt-install, where the VM just wouldn't restart after the disk had been formatted.