[ale] Virtual machine questions for public use machines

[JD]

PCI passthru of GPUs isn't easy.

VT-d is mandatory and kernel, BIOS, chipset support for it is as well.  The GPU
used matters too. The folks who tend to work on this stuff are gamers and use
higher-end GPUs, not the built-in graphics or $20 cards.  If Windows is the
guestOS, then manually installing the GPU drivers will be required. No setup.exe.

Lastly, I think shutting down the Windows VM is mostly a crash situation ...
from what I've read.

I've never attempted to do this.

On 01/26/2015 12:52 AM, Alex Carver wrote:
> On 2015-01-25 18:04, Horkan Smith wrote:
>> Alex, I don't have 'the' answer for you, but here's some thoughts:
> 
>> Do you have a preferred VM host software (kvm/qemu, virtualbox, vmware?)
> 
> No, no preferred software, just something that lets me run a GUI guest on a
> non-GUI host and also lets me kill the guest from the outside. After that
> anything is fine.
> 
>> 
>> I *think* to give the guest full screen control, you need PCI passthrough,
>> which looks to be hardware/ vm software dependent.  It's probably worth a
>> look, but don't be surprised if it's not available for your HW or SW.
> 
> I'll have to test this then, no problem.  The machines are old Dells with
> dual Xeons and a PCI graphics card though I'm not sure what. They're not
> going to have all the extra virtualization support that new processors do but
> it only has to support a very basic host and a fairly slim guest.
> 
>> 
>> If PCI passthrough won't work for you, then the next 'least stuff running
>> in linux' is probably something like directvnc (a framebuffer based vnc
>> client) running on the linux machine, pointing to a VNC server exported by
>> your VM software, or alternately by your guest OS.  There might be a
>> framebuffer version of a spice client or an rdesktop client, not sure.
>> Your mileage may vary, my experience w/ the majority of
>> direct-to-framebuffer apps is they're buggy and not worth it.
>> 
>> Last but not least, you could crank up an X-windows server with either no
>> window manager, or a small window manager configured to offer no
>> menus/escapes; then have your VM start in a full-screen window.  You can
>> turn off the alt-f1, alt-f2, etc in your xorg.conf file to help prevent
>> escapes.
>> 
>> With any of these options it's probably possible to point your linux
>> console to a serial port, if the machines have one.
>> 
>> FWIW, I run kvm most of the time 'cause I like the fact I don't have to
>> rebuild drivers when I switch kernels, and I like the command line
>> configurability.  I've used both vmware and virtualbox in the past and been
>> reasonably happy... for this application, I'd do a quick test to see if the
>> windows guest ran significantly better on different VM software.
> 
> I'll keep all of this in mind during testing.  It sounds like PCI passthrough
> is going to be my first stop since full screen control is exactly what I'd
> like to have, the illusion that the guest is the only OS on the system.
>