On Mon, 15 Jul 2002, Lukas Molzberger wrote:
> On Monday 15 July 2002 01:46, Mark Vojkovich wrote:
> > On Mon, 15 Jul 2002, Lukas Molzberger wrote:
> > > Hello,
> > > in recent years many people were talking about Linux on the desktop.
> > > However, before there is any real chance that this could happen a few
> > > fundamential problems in XFree must be solved. These are:
> > >
> > > 1. XFree is far too slow.
> >
> > No it isn't. Your apps are stupid or the drivers you are using
> > are under accelerated.
> I'm using Mozilla for example and I'm sure that its not that app that is slow
> here since I've compared it with Mozilla on WinXP. It may well be that the
> driver is under accelerated. I'm the i810 driver, but my old notebook with an
> smi chip was even slower. I know that the Nvidia driver is much faster but as
> far as I know the 2D part is still slower than under Windows even so it
> actually uses the same driver.
Netscape is much faster than Mozilla. I think it's just that some
design decisions in the X version of Mozilla, which is probably much
different than the Window's version, are suboptimal. Having seen enough
of GDI and XAA to do a comparison, I would say that XFree86 should be
able to clobber Windows in the intersection of their capabilities
given sufficient driver support.
The funny thing is that some people complain about this sort of
stuff with no evidence to the cause. It's not that it's impossible
to prove where the bottlenecks are either. Intel's VTune is available
for both Linux and Windows and can probably let you determine exactly
why the Mozilla is slower under Linux.
> I mean it has too much unnecessary complexety. I mean if the message system is
> only needed for the remote display feature then I'm really not sure if this
> feature is really worth the hassle. I've worked on two projects that
> contained an message system and in both cases it was a major problem that ate
> up a large chunk of the development time and it made the projects slow even
> if used on the same machine.
The protocol separates the client and server. This is a great security
advantage. X11's transport mechanisms have been stable for over a decade.
There is no hassle in using them and little if any developer time needed
to maintain it. OpenGL is direct rendered and skips the protocol and
this is very complex and prone to be buggy. It's good to avoid this
for 2D rendering and since the amount of data sent to the server for
2D (currently and likely in the future) is small there's not a performance
disadvantage to this.
> >
> > > 2d graphics drivers in users space while the 3d drivers are in kernel
> > > space?
> >
> > You are mistaken. 3D drivers are not in kernel space. OpenGL is in
> > user space in every OS I can think of.
> I'm pretty sure there is 3D driver part in kernel space. And I think it would
> be a good idea to also put the 2D part there. For example to have access to
> the interrupts.
The kernel driver is only to provide DMA buffers for user-space
apps to render to and also to provide mechanisms for user-space notification
of interrupts. It's bad to have too much stuff in the kernel. Graphics
devices are complicated and their drivers are complicated. When
you have a driver problem (and you will) and it's in user-space you can
still telnet in a reboot the machine. The more you move into the
kernel, the less likely you are to be able to do that. Increasing
the size and complexity of the kernel should avoided.
> >
> > As far as development being stuck, no, I don't think so. It's just
> > that the people who know enough about anything to get things done are
> > very few.
> True, but there are reasons for that.
>
The reason is that this stuff is difficult. We get so many college
kids that just learned C wanting to help. The help is appreciated, but
there are few things that they can do beside fix simple bugs and they
get discouraged. There's not much you can do about this. Changing
the project's mission statement doesn't make the work any easier. We
need people with years of experience either in graphics driver development
or in some other aspect of window system operation and that is hard
to come by. The fact of the matter is that dozens of new developers
with little or no window system experience are going to do little to
move the project forward.
Mark.
_______________________________________________
Xpert mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/xpert
