On 3/8/11 8:23 PM, Peter Robinson wrote: > The biggest issue I see in that is that most thin style corp > deployments now days are moving towards (or at least looking closely > at) a VDI style deployment and most of those techs don't support 3D > either. From a (very) cursory look at Windows 7, they appear to do Aero remoting by drawing the window itself on the machine running the app, and then compositing windows onto the display wherever the compositor happens to be running. Which is a pretty reasonable split; trying to remote all 3D rendering to the viewer can pretty easily swamp your bandwidth. The problem that then creates in a VDI kind of world is that the machine all your guests are running on now has to run a 3D renderer. I'm currently proof-of-concepting getting gnome-shell running with the software renderer, which would mean you could run it in kvm/vbox/etc, albeit with some CPU hit. After that if you want guest density you need to figure out a way to offload 3D rendering to the host's GPU. Thankfully, given how DRI2 and KMS work, I think that's mostly a matter of passing the right ioctls through to the host rather than trying to do PCI passthrough. All quite doable, just a lot of typing. Of course, I also expected gnome-shell on llvmpipe to be a one day hack, and it's turned into more like two weeks, so maybe I'm being optimistic about the complexity. - ajax -- devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxxx https://admin.fedoraproject.org/mailman/listinfo/devel
