Re: [AD] proposal: al_sleep()

[Elias Pschernig <elias@xxxxxxxxxx>]

On Fri, 2004-01-02 at 15:20, aj wrote:
> >
> >Yes, so it would make sense for yield_timeslice to release CPU under
> >windows and linux - even though it means, a complete timeslice is given
> >up. All the patch does is replace Sleep(0) with Sleep(1) - and it
> >achieves just that - CPU drops from 100% to 0%. If I'm in a loop where I
> >need all the CPU power, I wouldn't call yield_timeslice - therefore the
> >10 ms I effectively lose don't matter. If I'm in an idle loop, and call
> >yield_timeslice, I don't want the CPU to run with 100% - therefore
> >giving up the timeslice with Sleep(1) makes more sense than keeping to
> >run with 100% CPU with Sleep(0).
> 
> 
> is sleep(1)  really 1 ms ?       on which system ?
> sleep(1) can actually be more harmful than yield, due to thrashing.
> 

No, sleep(1) (small s) is 1 whole second.

> 
> if you know how much sleep you need then call sleep(>0) dont go trying to 
> change the meaning of yield.
> 

I just want 0% CPU when idle and all the CPU when not idle. For example
some code of a, let's say, platformer game:

while (1)
{
  while (game_frames < frames_ticker)
  {
    do_logic ();
  }
  if (rendered_frames < game_frames)
  {
    do_rendering ();
  }
  else
  {
    yield_timeslice ();
  }
}

When I had this on my old pentoim 200 MHz, it rarly got to call
yield_timeslice - and instead used 100% of CPU to update and display the
game. Now, I have a processor which can run the same game 10 times at
once. So it should use only 10% CPU.

Instead, what happens is, yield_timeslice () is called millions of times
heating up the CPU for no reason.

> 
> >And in Linux, my experience is that releasing CPU improves
> >responsiveness of my programs compared to running on 100% CPU. (I can't
> >explain why this would be the case though. Probably the linux scheduler
> >penalties apps being 100% busy, and the sched_yield doesn't change
> >anything in that.)
> 
> 
> 
> which linux scheduler ?    i dont know that much about the world of linux, 
> but aren't there different schedulers out there ?
> also, how can you say you get more responsiveness when its using less CPU.. 
> thats a complete contradiction.
> 

Yes, ideally, it would be so simple. But don't forget, there's more than
one thread running in Allegro, and there's a lot of background processes
running all the time in linux. I only noticed this behavior after some 
recent kernel updates - so yes, it is caused by a change in the linux
scheduler. And I can't really explain it as well.

-- 
Elias Pschernig <elias@xxxxxxxxxx>