[Asterisk-Dev] SIP/IAX repacking cpu cost?

[Andrew Kohlsmith]

On February 18, 2005 10:09 am, Kevin P. Fleming wrote:
> SIP-IAX on the other hand requires queuing and dequeueing frames between
> two different channel drivers. There is no way this can be as efficient
> as the frames staying inside a single channel driver (or protocol
> driver, in this case), no matter how much "tuning" may be done.

When converting the RTP stream to IAX2 audio frames, does Asterisk have to 
take the audio frame apart or transcode it?  i.e. if I have a gsm stream from 
a SIP phone to Asterisk and it's headed to an IAX2 destination on another 
Asterisk host, does Asterisk have to transcode from gsm to ulaw or slinear 
and then back to gsm, or is the SIP RTP gsm stream "compatible" with the gsm 
stream format in the IAX2 audio frames?

I'm asking because I'm looking to have a MaxTNT place SIP calls to a local 
asterisk box and then use IAX2 trunking to deliver the streams "enbloc" to a 
destination asterisk host on a remote network instead of having the MaxTNT 
stream the individual RTP streams directly to the far-end asterisk host.  

(other reasons to do this are to utilize the new jitter buffer to improve 
audio quality and to see whether it's better to use ulaw from the MaxTNT to 
the local * box and then trunk gsm to the far end or to let the MaxTNT do the 
ulaw<-->gsm conversion.)

I realize there's a performance hit to convert the audio from RTP to IAX2 
audio but I am curious as to whether Asterisk has to actually do any work to 
the audio itself or if it can just strip the RTP headers and prepend the IAX2 
audio headers.

-A.