PulseAudio as a sound output?

L. Rahyen research at science.su
Wed Oct 24 02:51:25 CDT 2007


On Monday October 8 2007 23:01, TheBlunderbuss wrote:
>  L. Rahyen wrote:
>  Please test with your uniprocessor and get back to us on it. Not all of us
> can have the top-of-the-line system. To all: stay away from Celerons! 128KB
> L2 isn't enough!

	Sorry for a big delay in reply. In practice it is very difficult for me to 
test anything on my server (because as I said it used for tasks related to 
forex and therefore I only can test something on weekends; this is sometimes 
is difficult too because currently I have a lot of work at weekends too 
unfortunately).
	I wasn't able to do full tests (compare different kernels, test all typical 
situations, etc.) because of lack of time and therefore I cannot post exact 
numbers to compare, sorry. But I post at least my impressions related to work 
of old uniprocessor system with 2.6.23 kernel.
	Well, everything is good. At least it seems to me that now everyone can 
forgot about sound issues when system is under heavy load: sound is clean 
even if I have a lot of working processes in background (I tested up to 10 - 
that's more than enough for uniprocessor system I think). I tried processes 
with and without heavy disk I/O - and sound was perfect.
	However, new scheduler isn't so perfect as it may seems at first glance. 
There is some issues with fairness anyway. And there is still no option to 
select frequencies higher than 1 KHz in vanilla kernel (this is sometimes 
useful for uniprocessor systems). I didn't run benchmarks so I cannot be sure 
but I think that my 2.6.22 kernel with some patches and hacks to the 
scheduler worked somewhat better, more smoothly under heavy load. Yes sound 
is clean and good but some applications doesn't run as smooth as I expect 
even with -20 niceness under heavy load (on my quad-core processor there is 
no such problem).
	Unfortunately I didn't test OpenGL (and games) in my uniprocessor system with 
2.6.23 kernel because of annoying bug in NVidia driver which happens sometime 
on this old machine: I have less than 1 FPS everywhere and only way to "fix" 
this is to reboot the machine (or reload the driver manually but this isn't 
working in all cases and require to shutdown X servers anyway). There was 
Sunday and about 20:00 GMT so I simply havn't time to test farther because 
after 21:00 GMT server should start its usual servicing.
	But even without any benchmarking I can say that 2.6.23 contain very big 
improvement in Linux scheduler. If you compare it with vanilla 2.6.22 or 
older (without any hacks and modifications to the scheduler) you will see 
that now many thing works perfectly even on uniprocessor system. As I have 
said before SMP systems have especially noticeable improvements.
	Now 2.6.23 is stable so everyone can easily try and test it. All major 
distribution should provide precompiled 2.6.23 kernels in near future.



More information about the wine-devel mailing list