Re: [Linux-ia64] Re: web page on O(1) scheduler

[Helge Hafting]

On Thu, May 22, 2003 at 05:18:02AM +1000, Duraid Madina wrote:
> Arjan van de Ven wrote:
> 
> >if you had spent the time you spent on this colorful graphic on reading
> >SUS or Posix about what sched_yield() means
> 
> Quoth the man page,
> 
> "A process can relinquish the processor voluntarily without blocking by 
> calling sched_yield. The process will then be moved to the end of the 
> queue for its static priority and a new process gets to run."
>
This assumes the implementation uses queues, one per 
priority level.
And even if it does, the process may be the only one with that
priority, making this a useless way of giving up "some time".
It'll still get rescheduled over and over and prevent
lower-priority processes from running.

 
> How you get from there to "I'm the least important thing in the system" 
> is, once again, beyond me. And even if that were a reasonable 
> interpretation of the word 'yield', you would still hope that more than 
> one CPU would get something to do if there was enough work to go around. 
> Agreed, "spinning" on sched_yield is a very naive way of doing 
> spinlocks. But that doesn't change the fact that it's a simple and 
> correct way. One would have hoped that calling sched_yield every few 
> million cycles wouldn't break the scheduler.

The way I understand it, the scheduler doesn't "break".  You just
get a lot of useless busy waiting.

Helge Hafting