On Thu, Sep 3, 2009 at 7:57 PM, Takashi Iwaitiwai@suse.de wrote:

But, *what* is actually the problem?  In your post, there was no clear description of the bug -- how is the behavior at all?

Problem is:

the channel driver uses Skype client, and uses an API call to make it do audio I/O to tcp sockets (to listening sockets of the channel driver itself).

*BUT* the Skype client insists to have a soundcard, probably for getting its timing from.

Actually interfacing with an actual soundcard is taxing badly the cpu (for dmix/dsnoop/plug overhead), takes irqs, and do not scale (you cannot have 30 Skype clients happily sharing the same soundcard).

So for "heavy duty" it uses the snd-dummy.

It works perfectly, and scales very well, with the Ubuntu 8.04 kernel, eg: a tickless 100HZ kernel. It works perfectly on Centos is uses the other snd drivers. It works perfectly on Centos with a compiled kernel like the Ubuntu 8.04 one. It works very bad (with glitches, accumulating delays, hiccups, etc) on Centos with the original Centos kernel (1000HZ, not tickless).

The ALSA unstable snapshot, and the first patch to it you sent me do not solves the problem.

Now I will try the last patches you sent me today and report here.

Thanks a lot for your efforts,

-giovanni