Grad project transition to masters thesis idea

Grad project transition to masters thesis idea

[Jeff Shanab]

I am starting a Grad class on graphics and planning on turning it into
my masters thesis.  I would like to write a openGL split graphics driver
for zen to allow sharing of accelerated graphics. (without copying large
amounts of data around)

Kind of xgl+xen in domain zero with each pane a seperate domain. I
envision a rotate left right between desktop and a rotate up down
between domains. Not really a cube anymore, but you get the point.   I
have been reading documents on xgl,compviz,xgtk,xegl and xen but I am
not sure if this is possible yet.

What I am thinking is that each opengl context will be "registered" with
the hypervisor and a virtual openGL driver provided in each domain will
serve out xgl for an x server to use. A domain switch is just a eye
point change or opengl context change, what is in the card stays there
between switches.

But... Am i correct in assuming that the driver in domain zero will have
to be made xen aware so that it can determine which domain interupts 
from code accesing the card belong to? What I am afraid is that since
the card has a GPU it is like a dual processor system and that there
might not be enough  access to the GPU, that it may need it's own
hypervisor . I don't really know how the GPU's work. Is the code running
in the main CPU or in the GPU, where is that line?

Suggestions, comments?

Re: Grad project transition to masters thesis idea

[Niraj Tolia]

Hi Jeff,

You might want to look at Xen-GL and Blink. They both virtualize
OpenGL for Xen VMs.

URLs: http://www.cs.toronto.edu/~andreslc/xen-gl/ and
http://www.diku.dk/hjemmesider/ansatte/jacobg/gfx/

Cheers,
Niraj

On 9/2/06, Jeff Shanab <jshanab@earthlink.net> wrote:
> I am starting a Grad class on graphics and planning on turning it into
> my masters thesis.  I would like to write a openGL split graphics driver
> for zen to allow sharing of accelerated graphics. (without copying large
> amounts of data around)
>
> Kind of xgl+xen in domain zero with each pane a seperate domain. I
> envision a rotate left right between desktop and a rotate up down
> between domains. Not really a cube anymore, but you get the point.   I
> have been reading documents on xgl,compviz,xgtk,xegl and xen but I am
> not sure if this is possible yet.
>
> What I am thinking is that each opengl context will be "registered" with
> the hypervisor and a virtual openGL driver provided in each domain will
> serve out xgl for an x server to use. A domain switch is just a eye
> point change or opengl context change, what is in the card stays there
> between switches.
>
> But... Am i correct in assuming that the driver in domain zero will have
> to be made xen aware so that it can determine which domain interupts
> from code accesing the card belong to? What I am afraid is that since
> the card has a GPU it is like a dual processor system and that there
> might not be enough  access to the GPU, that it may need it's own
> hypervisor . I don't really know how the GPU's work. Is the code running
> in the main CPU or in the GPU, where is that line?
>
> Suggestions, comments?
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>


-- 
http://www.cs.cmu.edu/~ntolia

Re: Grad project transition to masters thesis idea

[Mark Williamson]

> I am starting a Grad class on graphics and planning on turning it into
> my masters thesis.  I would like to write a openGL split graphics driver
> for zen to allow sharing of accelerated graphics. (without copying large
> amounts of data around)

Well, that'd be cool :-)

> Kind of xgl+xen in domain zero with each pane a seperate domain. I
> envision a rotate left right between desktop and a rotate up down
> between domains. Not really a cube anymore, but you get the point.   I
> have been reading documents on xgl,compviz,xgtk,xegl and xen but I am
> not sure if this is possible yet.

Kind of a hypercube then :-D  That'd be really cool.

> What I am thinking is that each opengl context will be "registered" with
> the hypervisor and a virtual openGL driver provided in each domain will
> serve out xgl for an x server to use. A domain switch is just a eye
> point change or opengl context change, what is in the card stays there
> between switches.

Jacob has done a bit of work on this in the form of Blink and also a previous 
project that source isn't available.  He might be able to suggest some 
reading...

> But... Am i correct in assuming that the driver in domain zero will have
> to be made xen aware so that it can determine which domain interupts
> from code accesing the card belong to?

You'd have to figure out some way of handling the multiplexing in domain 0.  
This doesn't necessarily have to modify the device driver, but *something* 
there needs to handle domains competing for resources, prevent them from 
abusing the screen (although preventing abuse might be more optional in a 
workstation setting).

> What I am afraid is that since 
> the card has a GPU it is like a dual processor system and that there
> might not be enough  access to the GPU, that it may need it's own
> hypervisor . I don't really know how the GPU's work. Is the code running
> in the main CPU or in the GPU, where is that line?

Another thing you might find interesting is that MS (and, I think, maybe 
NVidea) are looking at virtualisation-aware graphics hardware, but I'm not 
sure if there are any details available on this.

HTH,
Mark

-- 
Dave: Just a question. What use is a unicyle with no seat?  And no pedals!
Mark: To answer a question with a question: What use is a skateboard?
Dave: Skateboards have wheels.
Mark: My wheel has a wheel!