[Bf-committers] Re: [RFC] Re: Blender profiling-1 O16.2int

[Douglas Toltzman]

I agree with you, Hans, but I don't think "CPU starvation" is the correct
term.  CPU starvation is when cache misses and incorrect instruction
scheduling result in empty instruction pipelines which amounts to wasted
cycles.  Just my 2 cents.

On Wed, 27 Aug 2003, Hans Lambermont wrote:

> Ton Roosendaal wrote:
> 
> > Can someone translate this for me?
> 
> Basically it comes down to what is called 'busy waiting'. In a tight
> loop, perhaps with a sleep in it or not, that doesn't matter, a device
> value is read, over and over again, to detect whether it changes. This
> is called 'polling'. Too busy polling leaves no CPU time left for fi. X.
> The other approach (the better one) is to let the OS give you a signal
> when the device's value changed. This is done with a 'select' call.
> However, if you give this select call a too short timeout you get the
> symptoms of busy waiting again.
> 
> I do not know whether we use select at all here, but both select with
> too small timeouts and polling can cause the situation we're seeing.
> 
> 'hope this helps someone, I don't have the time to dig into it.
> 
> Hans
> -- 
> http://lambermont.webhop.org/   () ascii ribbon campaign - against HTML mail,
>                                 /\ vCards and proprietary formats
> _______________________________________________
> Bf-committers mailing list
> Bf-committers@blender.org
> http://www.blender.org/mailman/listinfo/bf-committers