On Mon, Oct 31, 2016 at 9:47 AM, Javier Celaya <javier.celaya@xxxxxxxxxxx> wrote:Hi,
El dom, 30-10-2016 a las 10:50 +0200, Sameeh Jubran escribió:
> Hi Fabio,
>
> I still suspect that the performance issue is caused by the mouse. As
> you can see in flexVDI's repository (
> https://github.com/flexVDI/qxl-dod/blob/master/qxldod/QxlDod.h) the
> function "EnablePointer" always returns FALSE,
Nope, check again. EnablePointer always returns FALSE for the VgaDevice
class, but always returns TRUE for the QxlDevice class. In particular,
returning TRUE is telling Windows "do not render the pointer, let theThat's right my bad, However it is always true in QxlDevice in flexVDI's repositorywhile it is always false in gitlab's repository.device render it itself", which means that in client mode, the spice
client is in charge of rendering the mouse pointer. This makes the
mouse run smoothly, but I am not sure how it is affected when vdagent
is not running.When it is false - as in the current gitlab repository - the driver keeps sendingunnecessary mouse info to the spice server which I believe is the cause forthe poor performance issues.So Fabio can you please try using the driver with EnablePointer - the one that'sin Qxl device class - set to TRUE as it is in flexVDI's repository?
Try by myself the FlexVDI driver. Yes, the problem is client/server mouse.
With FlexVDI, even with our agent installed client mouse is working.
Stopping the agent make even FlexVDI mouse slow (is easy to see the issue increasing
the network latency).
Not tried to change in our driver EnablePointer.
Frediano
> on the contrary to gitlab's repository (
> https://gitlab.com/spice/qxl-wddm-dod/blob/master/qxldod/QxlDod.h)
> where it always returns TRUE. I believe that setting it to false
> would solve the performance issue but would however cause another
> problem of which the mouse would be invisible when the vdagent isn't
> running.
>
> Can you please confirm that this solves the performance issue?
>
> If you need a compiled version of which "EnablePointer" returns
> always false, update me and I can supply it.
>
> On Fri, Oct 28, 2016 at 11:15 AM, Fabio Fantoni <
> fantonifabio@xxxxxxxxxx> wrote:
> > Il 28/10/2016 10:59, Christophe Fergeau ha scritto:
> > > On Thu, Oct 27, 2016 at 04:52:57PM +0200, Fabio Fantoni wrote:
> > > > Il 27/10/2016 12:03, Christophe Fergeau ha scritto:
> > > > > On Thu, Oct 27, 2016 at 10:40:47AM +0200, Fabio Fantoni
> > > > > wrote:
> > > > > > Is there any vdagent testing build with all newer patches
> > > > > > available
> > > > > > somewhere?
> > > > > https://copr.fedorainfracloud.org/coprs/teuf/spice-mingw/buil
> > > > > d/470179/
> > > > > has a totally untested build of a snapshot of
> > > > > https://cgit.freedesktop.org/spice/win32/vd_agent/commit/?id=
> > > > > debf1fe98148ddf1a44735449512bd4229b53213
> > > > >
> > > > > Christophe
> > > > Thanks for your reply.
> > > >
> > > > I tried it, with low resolution on notebook now it works
> > > > decently but on
> > > > thin clients with medium/high resolution is still too bad.
> > > Just to be clear, flexvdi builds were fine on the same hardware?
> > > medium/high resolution is 1920x1080 or similar?
> > >
> > > Christophe
> > with flexvdi was not perfect but significantly better
> > I did the tests with same server, vm and clients changing only the
> > qxl-dod driver and vdagent with the newer official ones.
> > For medium/high resolution of these tests was 1440x900 and
> > 1920x1080, 1920x1080 It is unfortunately a standard in the new
> > desktop "workstation" but with them on spice is difficult achieve
> > good performance, even with powerful thin client (for example,
> > recent octacore arm)
> > With flexvdi is usable but not very good on very basic use except
> > some cases with animation, video ecc..., with new official driver
> > is near unusable :(
> >
_______________________________________________ Spice-devel mailing list Spice-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/spice-devel