[Freeswitch-dev] Problem with RTP auto_flush code and hot_socket

[Brian West]

Please Open a jira... this is the second report I have heard of this  
issue.

/b

On Aug 6, 2009, at 4:29 PM, Cooper Xu wrote:

> Hi,
>
> In our load test, we found that the new RTP auto_flush and  
> hot_socket code after version 1.04 seems to cause voice quality  
> problem. We were stress testing Freeswitch with 250 channels using  
> codec G.711u and iLBC.  The RTP auto_flush appeared to be trigged  
> very frequently.
>
> Because our system is under load and socket buffer for RTP stream  
> sometimes accumulated a few packets before the channel can handle  
> it. When this happened, the RTP auto_flush and hot_socket mechanism  
> seemed sending those accumulated packets in a very short interval.  
> However after this burst, the RTP stream receiving will wait for  
> 80-100ms before receiving and sending another RTP packet. To find  
> out why this can happen, We looked at the RTP part of Freeswitch  
> code and found that inside the RTP read, during auto RTP flush, it  
> will call switch_core_timer_sync()  function. This will add 3 to the  
> timer reference count and the next switch_core_timer_next() call  
> will add another 1 to timer reference. This caused the next RTP  
> receiving being delayed 80-100ms. This 80-100ms delay for RTP  
> receiving caused 4-5 packets being accumulated in socket buffer  
> again. And it will soon trigger another auto_flush for RTP stream.  
> As the result, we saw a lot of short burst in the RTP stream, which  
> caused voice quality problem.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freeswitch.org/pipermail/freeswitch-dev/attachments/20090808/ac6b5e51/attachment.html