Re: [GnomeMeeting-devel-list] GnomeMeeting on 2.6

[Johnny Strom <jonny strom netikka fi>]

Hi all

I have very good news with the patch from Christian so
is gnomemeeting work better than ever.

I could run seti play two movies and do other stuff without
any disturbans in the sound.

So now I can say that GM with 2.6 is a perfect match at least for me :)


Thanks a lot Christian :)



Cheers Johnny





Christian Meder wrote:
> On Fri, 2004-01-30 at 11:41, Damien Sandras wrote:
>
> > Who is using the 2.6 kernel and is ready to test the latest changes with
> > it?
> >
> > Craig, given the fact that many packagers will compile on 2.4, what is
> > your suggestion?
> >
> > I tried on the 2.4 kernel to force the P_HAS_SEMAPHORES to 0 and I have a
> > bad sound quality when compiling, which I had not with P_HAS_SEMAPHORES
> > set to 1.
> >
> > The problem becomes complex. Having a bad sound quality is not
> > problematic. What was problematic on 2.6 was to have people having bad
> > sound quality as soon as they were moving a window.
> >
> > What are your comments?
>
>
> Hi,
>
> I took the gnomemeeting on 2.6 problem to linux-kernel around Christmas
> last year and I'm happily running gnomemeeting daily and stable since
> the discussion back then.
>
> Unfortunately I switched my laptop from a PIII 450Mhz to PIV 2GHz where
> the problem is less obvious. But still even on the 2 GHz laptop the
> sound gets distorted with slight load on the machine in the
> gnomemeeting-cvs Debian default packaging.
>
> BTW it's really unlikely that the problem has anything to do with audio
> drivers as the problem persisted whether I used OSS or ALSA on a
> mastro2e chipset (450 MHz) or ALSA on the intel8x0 (2GHz).
>
> Ingo Molnar (the Linux scheduler maintainer) pointed out in December
> that the sched_yield usage in pwlib is broken. He suggested to just
> nanosleep for 1 timeslice instead.
>
> I tested two approaches:
>
> * compile a shared library with sched_yield defined to a noop and run
> gnomemeeting via library preloading with this shared library, called the
> romp hack in a paper by David Mosberger-Tang
>
> * substitute the Yield function in pwlib with nanosleep
>
> The first approach seems to hang the GUI and slow down the GUI update
> rate resulting in a very sluggish gnomemeeting feeling. Nevertheless
> audio quality is improved but at a GUI usability cost.
>
> The second approach is what I'm using since December and audio quality
> is rock stable even under heavy load for me.
>
> --- pwlib-cvs-20040128.CVS.orig/src/ptlib/unix/tlibthrd.cxx
> +++ pwlib-cvs-20040128.CVS/src/ptlib/unix/tlibthrd.cxx
> @@ -1057,7 +1057,8 @@
>                                                                                 
>  void PThread::Yield()
>  {
> -  sched_yield();
> +  struct timespec timer = { 0, 1 };
> +  nanosleep (&timer, NULL);
>  }
>
>
> Note that I didn't try to play around with the P_HAS_SEMAPHORES because
> I just wanted to fix the sched_yield usage without changing much else.
>
> So in summary after compiling a patched pwlib my gnomemeeting problems
> where gone. Without the patched pwlib gnomemeeting was unusable even
> with the slightest load on the machine be it a 450 MHz or 2 Ghz
> processor.
>
> I'm interested in experiences from others.
>
>
>
>
> 			Christian