[asterisk-users] Latency in ConfBridge conferences

[Nicholas Barnes]

Hi all,

I've done a basic install of 10.1.2 to have a play with the new 
ConfBridge application and have noticed high latency when in a 
conference. It's to the order of 900ms or so which is just too much for 
a conference to work well.

I can account for about 120ms of that latency, but not the rest.

With two SIP handsets in the conference, worst case latency from handset 
to handset (Handset --> Local Asterisk --> Gateway Asterisk --> PSTN --> 
Gateway Asterisk --> Conference Asterisk --> Gateway Asterisk --> PSTN 
--> Gateway Asterisk --> Local Asterisk --> Handset) is measured at 62ms 
which leaves 780ms or so latency created by the conference bridge.

The box I'm testing ConfBridge on is a CentOS 5.7 x86_64 box with a Xeon 
E31220 and loads of RAM. Whilst a two party conference is in progress, 
'top' shows a load average of 0.05 and that Asterisk is using 0.5% CPU. 
Assuming the ConfBridge works in Slin, 'core show translation' table 
shows a 6ms (which does strike me as rather high - I get much lower 
figures on much older hardware). There is no jitter buffer and the 
mixing interval is 10ms. I'm pretty sure there are no bottlenecks in the 
hardware!!

I was wondering if anybody else had experienced the same thing and/or 
whether this is normal behaviour. Can I work it faster? Any ideas, or 
should I just accept this?

Nicholas.