[pulseaudio-discuss] [PATCH] core: New LIFO style flist implementation

[Jyri Sarha]

On Mon, 23 Aug 2010, Shouqun Liu wrote:

> Hi Jyri,
> I have applied the patch, but makes no effort to improve the performance,
> the CPU usage is still high :)
>

Well, I originally created the patch to avoid pulseaudio getting stuck
while waiting swapped out memblocks on Nokia N900. After applying the
patch I also noticed that it slightly improved the performance in some
usecases. I think that was because of better CPU cache hit ratio, but I
guess you need to have a lot of concurrent streams or some audio filtering
ongoing on a limited memory-bandwidth device like N900 to notice the
difference.

> I'm still confusing about this issue, there is no such problem in my PC, but
> its critical in my netbook.
> btw, the PA version on PC and netbook are both latest 0.9.21 release.
>

Do you have a constant high load on your netbook or do you see it for
instance while PA is rewinding streams? You should be able to decrease
the load spikes from rewinding by shortening the ALSA DMA buffer.

Do your streams have the same sample-rate as your default sink?
If no, relaxing your sample-rate-converter settings should help (or 
configuring your sink to match the samplerate of your stremas).

Cheers,
   Jyri