Re: Responsiveness of 2.6.0-Test9

[Nick Piggin]

Erwin Telser wrote:

> > Erwin Telser wrote:
> >
> > > Are there things to observe when switching from 2.4 to 2.6, if I want the
> > > same responsiveness? I'm asking because I' ve made the following
> > > observation. (I always had the feeling, 2.4 seems to faster, but this is
> > > the first time it' s very obvious)
> > >
> > > I' ve connected two monitors on a Matrox G550. One runs with DRI the other
> > > one doesn't (not possible with the current driver). The Monitor with DRI
> > > I'm using as a TV Set, watching movies with xawtv. (With a bttv 878 tuner
> > > card).
> > >
> > > Now with the 2.4.22 kernel (preemptible patch aplied) I can play the
> > > little game kbounce on the other monitor, without to notice any slowdown,
> > > no matter, whether xawtv is running or not.
> > >
> > > But with the 2.6 Kernel (compiled with preemptible option) the bouncing
> > > balls slow down considerably, as soon as I move the mouse.
> > >
> > > I know the whole thing is a little foolish. But anyway, are there some
> > > tricks to get the same responsiveness?
> > Hi,
> > Your report is not foolish at all ;)
> > Please make sure your X process is at default static priority, 0.
> > Report back if you are still having problems. What sort of processor
> > do you have? Is your 2.6 video driver running with the same sort of
> > acceleration as your 2.4 one (do you get the same frame rate in glxgears)?
>
> I installed X from scratch and didn't edit any start script. so I assume, that 
> it runs with priority 0. The processor is an Intel P4 2000MHz in an ASUS 
> P4B266 Board. If I compare the frames per second with glxgears there is a 
> slight advantage for 2.4

renice 0 `pidof X`

> 2.4: 
> with DRI 358 - 364
> without DRI 181 - 198
>
> 2.6
> with DRI 350 - 362
> without DRI 180 - 187
>
> An even better way to show the differenze between 2.4 and 2.6 is to change the 
> size of the xawtv window. If I do that, the slowdown is quite significant. 
> But it happens only with xawtv. If I do the same with konqueror or MPlayer, 
> nothing happens. A very strange thing are the values I get from ksysguard
>
> Kernel 2.4, glxgears
> user load: 3% - 4%
> system load: 95%
> IRQ per second: ~100
>
> Kernel 2.6, glxgears
> user load: 15%
> system load: 85%
> IRQ per second: ~1100 
>
> Kernel 2.4, xawtv
> user load: 2%
> system load: 0%
> IRQ per second: ~100
>
> Kernel 2.6, xawtv
> user load: 9%
> system load: 0%
> IRQ per second: ~1060 
>
> Kernel 2.4, MPlayer
> user load: 35%
> system load: 2%
> IRQ per second: ~1170
>
> Kernel 2.6, MPlayer
> user load: 42%
> system load: 2%
> IRQ per second: ~2130
>
> Something is causing a high IRQ load with kernel 2.6. Probably an APIC 
> problem?

It is the timer interrupt. Now happens 1000 per second instead of 100.
This might just be the reason why you are seeing a different balance, the
collection of cpu statistics is more precise.


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/