Re: 2.5.74-mm1

[Nick Piggin <piggin@cyberone.com.au>]

Davide Libenzi wrote:

>On Tue, 8 Jul 2003, Nick Piggin wrote:
>
>
>>I agree some people have some inflated ideas about a desktop workload,
>>but I'd just point out that if your mp3 player was using say 2% CPU,
>>it should be able to preempt the make soon after it becomes runnable,
>>and not have to wait at the end of the queue. It would become a CPU
>>hog itself if you had 48 other processes running though.
>>
>
>This is clearly true, actually even more since player usually suck way
>less than 2% of the CPU. If no video is involved, they simply do a write()
>to /dev/dsp and then they sync by calling GETOSPACE and sleeping in the
>"hope" to be wake up almost in time. I never said that the scheduler
>should not be fixed. It definitely has to.
>

Well, yeah, I can run xmms on my 2xCPU system with 32 processes in
an infinite loop, and 32 in an infinite loop of fork+wait. No skipping.
So maybe gcc gets its priority elevated too much due to the small
amount of waiting it does.