I've started running some stress tests of both SIP signaling and RTP handling in Asterisk 1.4 and Trunk. In our telecomm routing project we needed a SIP front end to our route selection server and the choice became Asterisk. However, we need to be able to push at least 400-500 calls/second through Asterisk to any given Griffin server. To do this is going to require some work on the Asterisk side of things (as well as a herculean effort on our Griffin development team).
We've started by implementing a "300 Multiple choices" response in Asterisk (still no inbound request support yet) and then intentionally breaking it to support the horrible format that our switch uses. After Griffin is in production in our environment for a while, I'll fix the 300 message and we'll commit that to the project proper.
Today we hit roughly 9 million calls routed in 48 hours (or 50 calls/second sustained). Now, it's important to note that these are very short calls, and have no RTP. However, the SIP signaling is being done very efficiently.
My next set of tests involves setting up servers to push the limits of Asterisk concurrent call handling (with RTP this time). Our hope is that interrupt coalescence will improve concurrent capacity.
As soon as I have better test results that are in a more formal format, I'll be happy to share them.