dracut hostonly and cloud images

[Josh Boyer]

On Mon, Apr 8, 2013 at 12:06 PM, Matthew Miller
<mattdm at fedoraproject.org> wrote:
> On Mon, Apr 08, 2013 at 11:08:03AM -0400, Josh Boyer wrote:
>> > We're going to produce official Fedora Cloud images which should run on
>> > Amazon EC2, and various open source cloud infrastructure -- OpenStack,
>> > Eucalyptus, CloudStack, OpenNebula; with both KVM and Xen -- and probably
>> > also under VMware and VirtualBox.
>>
>> Can you elaborate on those last two?  Are you intending to generate
>> images explicitly for VMWare and VirtualBox, or do you expect the cloud
>> images to just run there?  If the former... please explain further.
>
> The latter, definitely. And of those two, with priority to VirtualBox
> because it's a) open source (extensions notwithstanding) and b) heavily used
> on Mac laptops where Fedora is a common guest. (The cloud images isn't
> intended to be a generic guest image, but it's handy if it happens to work.)

VirtualBox might be open source, but it is consistently broken.  The
VMWare hypervisor is closed source, but they actually got most of their
kernel drivers upstream now (with the major exception of the hostfs one).
We build and ship those drivers in the Fedora kernel.

I'm not pretending that vbox isn't the more popular of the two, but from
a kernel support standpoint it is certainly worse.  We simply don't look
at anything reported in VirtualBox unless it can be recreated in KVM or
on real hardware.  Mostly just an FYI.

>> In both cases, is there a reason you're excluding Hyper-V?
>
> No, no particular reason; I just left it out of the list. I can't really
> test it myself, and the same lower-priority-since-it's-not-open applies. And
> in all cases we're certainly (and obviosuly I hope) not packaging up guest
> driver / toolkit software that isn't free software.

Like VMWare, most of the hyperv kernel drivers are in the upstream kernel
now.  We also ship those in the Fedora kernel.  Though of the three, I'd
certainly agree it's the lowest on the list and none of the kernel team
has an environment to actually test anything.

josh