[Live-devel] RE: RTP packet rate smoothing?

[Karl Gierach]

Ross,

I have a seemingly related problem where I experience CPU thrashing.
I am streaming real time audio recorded from a telephony interface.
The audio is fed into a buffer from the telephony interface at a
constant real-time rate, and I have code which integrates with the
live555 framework and reads the audio from this buffer.

I am finding that, very quickly, my buffer becomes empty.
I have traced a symptom of the problem down to this location of code in 
MultiFramedRTPSink::sendPacketIfNecessary(). The computed value of
"uSecondsToGo" gradually becomes negative or 0, and my CPU begins to
thrash a bit as the active live555 thread continues to query the buffer
for data even though it is empty.


    if (fNextSendTime.tv_sec < timeNow.tv_sec) {
      uSecondsToGo = 0; // prevents integer underflow if too far behind
    } else {
      uSecondsToGo = (fNextSendTime.tv_sec - timeNow.tv_sec)*1000000
	+ (fNextSendTime.tv_usec - timeNow.tv_usec);
    }

        // karl begin
	  if( uSecondsToGo < 0 )
	  {
        uSecondsToGo = 20000; // std packet length - 20 ms
	  }
        // karl end

I see that my first attempt to fix this is similar to what you have
proposed below. In my particular situation, should I stick with this
solution or would you recommend something else?

Thanks,
-Karl

>>Any suggestions on to spread out the packets that make up a given
>>frame so they are uniformly distributed over the frame time
>>(durationInMicroSeconds).
>
>A better (or at least easier) solution would be to ensure that 
>there's at least a minimum delay between successive outgoing packets, 
>so as to not exceed the overall bit rate of the network.
>
>I.e., add the following statement just before the call to 
>"scheduleDelayedTask()" in "MultiFramedRTPSink.cpp":
>         if (uSecondsToGo < minInterPacketGap) uSecondsToGo = 
>minInterPacketGap;