[pulseaudio-discuss] Questions, questions and questions!

Max Kempter kxam at gmx.net
Thu Nov 30 04:38:37 PST 2006


Now I try to phrase my questions in a better way.

1. Hardware requirements for the server?
Pierre wrote: No idea. It all depends on what you want the server to handle.

I wrote:   For a system with 20 Server (each per host). For each Host it should be
possilbe to send to the others simultaneously. So in the worst case I have
19*20 active channels.
 I only want to send voice I have a data rate for one channel about
64kb; for 19*20channels=380 channels; if I send each 3 ms a packet the data rate will go up to 92,1 Mbit/s (include the Ehternet-, IP-,UDP-, RTP-Header and 5 per cent RTCP- Traffic).

Cj wrote:  Does each host transmit 19 *different* streams? If not, you really should belooking at multicast, which would give you only 20 streams on the wire.
3 ms sounds pretty over the top. Do you really need that? The standard for
realtime audio streaming is usually 20ms (ie. 50Hz), which would give you
around 25 Mbit/s for 380 streams and only 1.3 Mbit/s for 20 multicast

I write: 

I really need this 3 ms latency, because the end-to-end transmission should be less than 10 ms.
I work with my own LAN there is no problem with the network capacity, I don`t use a public net.
My ambition is to have up to 20 hosts/servers who can comunicate with any. As keyboard I will use a desk with 20 buttons. Behind each button is the order to  send to the Host with the number of the button (with the something like "pacat -p -s").
Each host transmit 19 equal streams in the worst case scenario(but this is very very improbable). Normally there are 4 or 5 active simultaneously
Normally each host send only to one or two other host, this is why I think it`s not necassary to use multicast. The worst case is to use 380 channels; it`s very improbable. Because normally each person have only 10 fingers and can only push 10 buttons. 

2. Optimise the Debian kernel for PA? 

I wrote:  Is it a good a idea to use Audio kernel with RealTime Preemption, if I want a low latency system?

Pierre wrote: Pulse does use any of the realtime syscalls, so I don't think it will get any performance gain.

I write: Is there an other possibility to optimize the kernel for pulse audio or should I optimise it for AlSA?

3. A Problem with Alsa-Source!

When I try to load my source, I can not adjust the sample rate with:


>>>load-module module-alsa-source rate=8000 source_name=mic
alsa-util.c: device doesn't support 8000 Hz, changed to 48000 Hz.


But if I try to record with arecord, there is no problem to set the sample rate to 8000Hz?

4. Could you explain the context between latency of alsa-modules to the fragments and fragment_size value? Is it like to set the buffer-size (eg with arecord)? 
Tell me more than, if you decrease fragments and fragment_size you decrease latency, please.  

5. Is there an advantage to use add-autoload-sink/source instead load-module sink/source? When I load the sinks/sources with autoload I don`t have access to the sinks/sources (eg with play-file).

I`m very thankful about your patience.



"Ein Herz für Kinder" - Ihre Spende hilft! Aktion: www.deutschlandsegelt.de
Unser Dankeschön: Ihr Name auf dem Segel der 1. deutschen America's Cup-Yacht!

More information about the pulseaudio-discuss mailing list