Re: O(1) scheduler starvation

[William Lee Irwin III]

On Wed, Jun 18, 2003 at 02:04:45PM +0200, Mike Galbraith wrote:
> I've got thud licked (without the restricting sleep to time_slice), 
> test-starve works right as well, and interactivity is up.  Tasks waking 
> each other in a loop is a bitch and a half though, and need to be beaten 
> about the head and shoulders.  Going to a synchronous wakeup for pipes 
> (talking stock kernel now) cures irman process_load's ability to starve... 
> IFF you're running it from a vt.  If you're in an xterm, it'll still climb 
> up from the bottom (only place where it can't starve anybody) and starve 
> via pass-the-baton wakeup DoS.  That will/does take the joy out of using 
> xmms.  If xmms didn't use multiple threads, it'd be much worse... right 
> now, you'll lose eye-candy [cpu hungry visualization stuff] before you lose 
> sound [at next song].

That's great. I'd love to see the patch.

-- wli